Decoding Perinatal Mental Health: Investigating Protective and Risk Factors and Predictive Insights for Aboriginal Perinatal Mental Health through Explainable Machine Learning

Decoding Perinatal Mental Health: Investigating Protective and Risk Factors and Predictive Insights for Aboriginal Perinatal Mental Health through Explainable Machine Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Decoding Perinatal Mental Health: Investigating Protective and Risk Factors and Predictive Insights for Aboriginal Perinatal Mental Health through Explainable Machine Learning Guanjin Wang, Hachem Bennamoun, Wai Hang Kwok, Jenny Paola Ortega Quimbayo, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4687233/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Perinatal mental health significantly affects mothers, infants, and families. Despite their resilience and strengths, Aboriginal mothers experience disproportionate physical and mental health disparities. These result from historical and ongoing impacts of colonization and the resultant complex trauma. Conventional approaches to perinatal care present many barriers for Aboriginal mothers who frequently feel disengaged, apprehensive and unsafe. Current score-based risk-screening practices that algorithmically drive referrals, further ingrain fears including culturally biased judgments and child removal. The Baby Coming You Ready (BCYR) model of care centred around a digitised, holistic, strengths-based assessment, was co-designed to address these barriers. The recent successful pilot demonstrated BCYR effectively replaced all current risk-based screens. However, many professionals disproportionately rely on psychological risk scores, overlooking the contextual circumstances of Aboriginal mothers, their cultural strengths and mitigating protective factors. Methods To address this singular reliance screening psychometrics whilst supporting strengthened culturally considered clinical assessment, we propose a culturally sensitive eXplainable AI (XAI) solution. It combines XAI with Aboriginal lived experience, knowledge and wisdom to generate a clinical prediction model to support professionals and Aboriginal mothers being screened. The XAI solution can identify, prioritise, and weigh both maternal protective strengths and risk factors, quantify their relative impacts on perinatal mental-health and well-being at both group and individual levels. Results Different machine learning algorithms, including Random Forest, K-nearest neighbour, and support vector machine, alongside glassbox Explainable Boosting Machine (EBM) models, were trained on the real life de-identified data generated during the BCYR pilot. Additionally, XAI techniques like SHAP and LIME are utilised for interpretability on black box models. Results show the EBM model demonstrates superior performance in prediction, with an accuracy of 0.849, F1 score of 0.771 and AUC of 0.821. Global explanations across the entire dataset and local explanations for individual cases, achieved through different methods, were compared and showed similar and stable results. Conclusions This study demonstrated the potential for this XAI solution to enhance professionals' capability in culturally responsive clinical reasoning in perinatal mental-health screening to improve experience and strengthen outcomes for Aboriginal women. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 1. Background Perinatal depression and anxiety negatively impact the health and well-being of mothers and babies, and disrupts maternal /infant bonding [ 1 ]. Recent studies highlighted the significant association between perinatal depression and adverse outcomes, including suicidal behaviours and self-harm thoughts during and after pregnancy. AR Mitchell, H Gordon, A Lindquist, SP Walker, CS Homer, A Middleton, CA Cluver, S Tong and R Hastie [ 2 ] emphasised the increased risk of preterm birth, stillbirth, and suicide associated with perinatal depression. Furthermore, AD Hummel, K Ronen, A Bhat, B Wandika, EM Choo, L Osborn, M Batra, J Kinuthia, M Kumar and JA Unger [ 3 ] highlighted the association of perinatal depression with adverse infant outcomes such as preterm birth, intrauterine growth restriction, and low birth weight. The loss of an infant's mother through suicide profoundly impacts the infant's social and emotional well-being [ 4 ]. Many Aboriginal women experience strong social and emotional well-being and have flourishing infants and families. However, at a population level, too many Aboriginal women face the increased risk of triggering or worsening depression/anxiety directly resulting from the enduring challenges, barriers and adversities from colonisation, cultural disruption and past and present policies such as the Stolen Generations . These include poverty, racism, intergenerational and complex trauma, racism, cultural bias, loss of cultural identity and other inequities [ 5 , 6 ]. A systematic review by S Owais, M Faltyn, AV Johnson, C Gabel, B Downey, N Kates and RJ Van Lieshout [ 7 ] indicated Aboriginal women face a 38% higher chance of experiencing depression, are 79% more susceptible to mental-health problems during pregnancy, and 30% more likely to endure mental health complexities post giving birth. A Western Australian study revealed that between 1997 to 2013, one in three Aboriginal babies were born to mothers who sought hospital care for mental-health issues related to substance abuse, depression and/or anxiety [ 8 ]. Despite routine screening for perinatal depression and anxiety in Australia for over 20 years, the gap in Aboriginal mothers’ and infants’ health and wellbeing remains unacceptable across all key indicators. This is evident in disproportionately higher rates of premature births, low-birth weight babies and child removal [ 5 , 6 ]. Conventional health systems approach to antenatal/postnatal care and screening are often culturally insensitive and retraumatising for Aboriginal women [ 9 ]. Risk-focused perinatal screens and assessments with Aboriginal women frequently exacerbate feelings of alienation and disengagement from potentially supportive care [ 10 ]. There is an urgent need for, culturally safe and effective trauma-aware and healing informed screening for social, emotional, mental-health and well-being that includes relevant supportive and strength-based follow-up care for Aboriginal mothers. The Baby Coming You Ready (BCYR) program [ 10 ], was co-designed to overcome these barriers and challenges faced by Aboriginal parents during their perinatal care. The BCYR Program centres around a digitised, strengths-based, culturally safe, and holistic perinatal assessment that incorporates all seven elements of Aboriginal peoples’ social and emotional well-being [ 11 ]. The assessment uses iPads with touchscreen images depicting common strengths, worries and past and present occurrences. Aboriginal voice-overs on each slide guide the reflective engagement between the mother and her midwife or health professional. Mothers choose images they relate to while engaging in self-reflection, creating their own personalised story, then prioritise their strengths and concerns before designing their own solutions. Currently the BCYR program is operationalised as a model of care in all six pilot sites in Western Australia (WA), and effectively replaces all current required screens for mental-health, family and domestic violence (FDV), tobacco, alcohol and other drugs (AOD). Whilst the successful pilot demonstrated increased trust, engagement, honest disclosure and self-directed management plans, it found that some midwives and managers lacked confidence in culturally considered assessments [ 10 ]. They continued to overly rely on the embedded risk-screen score (Kessler-5 distress scale) which undermines the trauma-aware and healing-informed approach. It weakens the potential of the BCYR assessment and health clinicians to effectively support Aboriginal mothers’ strengths and self-determined solutions, and ultimately contribute to improve maternal/infant outcomes and public health cost savings. Over the last decade, with the advancements in digital health and computational capabilities, numerous studies have focused on technology-based approaches to advance support for perinatal mental health [ 12 , 13 ]. This includes the development of Artificial Intelligence (AI)-based models, particularly machine learning and deep learning-based models, for predicting perinatal mental-health conditions [ 14 – 18 ]. These AI-assisted decision-support approaches in mainstream populations have potential to facilitate early and accurate detection of depression, thereby enabling timely interventions. They may also assist in identifying patterns that might be overlooked in manual assessments, leading to improved clinical efficiency and outcomes [ 19 ]. To mitigate the identified service gap in Aboriginal-focused perinatal mental health screening and care, our aim is to integrate Aboriginal ways of being and knowing with AI technology to develop a culturally considered, strengths-focused clinical prediction model. Specifically, our approach introduces eXplainable Artificial Intelligence (XAI) technologies, including glassbox models and post-hoc agnostic models , into our solution to elucidate the rationale behind AI-driven decision-making. This approach can identify, prioritise, and weigh both maternal protective and risk factors, and quantify their relative impacts on perinatal mental health and well-being outcomes at both group and individual levels. It is anticipated that through the ongoing Aboriginal co-design and cross-disciplinary process, this AI-driven clinical prediction model will support healthcare professionals and Aboriginal women in developing an understanding of and deepened trust in the model's predictions. It will also enhance health professionals’ confidence in holistic, culturally contextualised clinical decision-making and minimise the cultural biases held by clinicians, and enhance trust and engagement with Aboriginal mothers toward their health professionals’ supportive care and in their perinatal health service [ 20 ]. The remainder of this paper is organized as follows. Section 2 briefly introduces existing research related to studies using AI techniques for the Aboriginal population and XAI techniques on mental health issues. Section 3 describes the dataset adopted, the machine learning models, and the XAI techniques used for the predictive modelling. The results, along with discussions, are presented in Section 4. Section 5 provides the final discussion and conclusion and suggests future work. 2. Related Work This section contains two parts. Section 2.1 provides an overview of AI development to support the health care of the Aboriginal population. Section 2.2 reviews recent research that utilizes XAI techniques in the mental health field, with a specific focus on structured tabular data, which is the same data type encountered in this work. 2.1. AI Applications in Aboriginal Population Healthcare While there is increasing interest in utilising AI techniques to address diverse healthcare challenges, there remains a notable scarcity of studies within Aboriginal communities, as evidenced by the limited number of studies conducted in recent years. One study aimed to classify ear diseases through the analysis of otoscopic images collected between 2010 and 2018 from Aboriginal children in the Northern Territory, Australia [ 21 ]. Otolaryngologists analysed and classified the images to label the diseases identified in each screening. Subsequently, this labelled dataset was used to train an image classification algorithm employing deep and transfer learning techniques, achieving an accuracy ranging from 74.4–92.8% [ 21 ]. J Scheetz, D Koca, M McGuinness, E Holloway, Z Tan, Z Zhu, R O’day, S Sandhu, RJ MacIsaac and C Gilfillan [ 22 ] developed a convolutional neural network to detect the presence of diabetic retinopathy in participants, including patients from three Aboriginal Medical Services clinics. Overall, the results showed an accuracy of 96.6% in predicting diabetic retinopathy and 87.7% in identifying retinas without the disease. S Chando, JC Craig, L Burgess, S Sherriff, A Purcell, H Gunasekera, S Banks, N Smith, E Banks and S Woolfenden [ 23 ] implemented a multimodal logistic regression model to assess risks in the development and well-being of Aboriginal children. The data were obtained through surveys conducted by Health Services in New South Wales, which included demographic, physical, family, and social factors influencing child health development and well-being for model construction. The study revealed that 32% of the subjects were at high developmental risk, 28% at moderate risk, and 40% at low or no developmental risk. On the other hand, little has been accomplished using AI techniques to support health improvements with Aboriginal populations. A recent systematic review of 13 studies on cardiometabolic diseases in primary healthcare highlighted the notable absence of studies employing AI to address heart-related and metabolic diseases within Aboriginal communities [ 24 ]. However, existing machine learning models, typically trained on data from the general population, may not be suitable for the Aboriginal populations. Historically, Aboriginal peoples have been subjected to mainstream health related ‘solutions’ with little to no consideration given to their cultural and social determinants of health, or poor access to primary health care and potential biases in the data. Furthermore, many existing AI-assisted decision-support approaches prioritise enhancing prediction accuracy, often leading to increased model complexity due to intricate mathematical modelling [ 25 ]. However, these approaches often lack transparency in rationales and explanations for the decision-making. Applying existing AI models in Aboriginal communities may have grave risks of data biases, objectification and further exacerbation of health inequality. We propose co-designing an AI model by integrating Aboriginal lived experiences, knowledge and wisdom with professionals’ experiences and XAI techniques. This approach will ensure the development and deployment of AI technologies align with Aboriginal community desires, needs expectations, while empowering clinicians to make culturally informed and safe, evidence-based decisions via trustworthy AI tools. 2.2. XAI for Mental Health with Tabular Data Many AI models are often regarded as “black boxes” due to their lack of interpretability and transparency [ 26 ]. To address this issue, XAI has emerged as a promising area, offering easy-to-understand explanations of the predictions and decision-making processes of AI and ML algorithms [ 27 , 28 ]. In mental health applications, understanding the inner workings of AI models becomes crucial. Decision-making in mental health can be highly complex and influenced by various circumstances, making transparency essential for stakeholders [ 29 ]. N Jothi and W Husain [ 30 ] implemented the Shapley value under the proposed method as the feature selection of the data mining classifier on mental health data to predict generalised anxiety disorder among women. This enhanced prediction solution aimed to provide help to any women perceived to be suffering from the disorder. M Ikram, NF Shaikh, JK Vishwanatha and U Sambamoorthi [ 31 ] applied eXtreme Gradient Boosting and SHapley Additive exPlanations (SHAP) on survey data, including demographic and socio-economic information, to identify predictors of poor mental health and explain their associations with poor mental health. AU Hussna, II Trisha, IJ Ritun and MGR Alam [ 32 ] investigated the impact on students’ mental health during the COVID-19 pandemic. They identified the top influential variables impacting model output based on SHAP values, including gender, sleep disturbance, perceived discrimination, sitting hours, and stress perception. Similarly, V Adarsh and GR Gangadharan [ 33 ] employed SHAP and Local Interpretable Model-agnostic Explanations (LIME) to evaluate IT workers at risk of mental illness, detect early signs of depressive symptoms and use them for training the prediction model. By leveraging common explainable techniques like SHAP and LIME, we can not only enhance predictive precision but also provide clear justifications on the decision-making process, which are imperative for informed clinical decision-making and the formulation of therapeutic interventions [ 34 ]. This aspect also becomes crucial for fostering trust and ensuring the deployment of AI systems that are suitably sensitive and effective within the mental health domain and targeted groups. 3. Methods 3.1. Data Source The dataset is drawn from deidentified data generated by the digital assessments that were completed during the WA BCYR pilot [ 35 ]. This culturally safe and strengths-based perinatal mental-health and social and emotional well-being assessment generates an automated clinical event summary (PDF) which serves as an individualised follow-up management plan for the mother and health professional. The de-identified dataset emerged from 293 Aboriginal pilot participants (mothers) between September 9th, 2021, and June 16th, 2023, from one of eight perinatal health-care sites in six services across metropolitan Perth and regional WA. Each participant’s data includes 345 variables recorded across different inquiry domains, encompassing strengths and culturally protective factors, common life events, worries, quality of relationships, childhood experiences, family and domestic violence (FDV), tobacco and AOD use. The indicator for maternal distress is derived from the Kessler-5 item psychological distress scale (K5) [ 36 ]. 3.2. Data pre-processing Five records were excluded from the dataset due to missing information regarding the perinatal mental health condition indicator. On the K5 assessment instrument, each item was measured using a five-point scale ranging from 1 to 5. All K5 items are negatively keyed. A score of between 5 to 25 was possible, less than 12 was considered low risk (0); a score of 12 or higher was considered high risk (1) [ 37 ]. A comprehensive pre-processing and data cleaning procedure was followed by feature selection process that utilised the variable importance measure computed by the random forest algorithm. The algorithm was configured with 500 trees, and the “mtry” parameter was set to the square root of the total number of variables, rounded down to the nearest integer. Initially, the top 31 most significant variables were selected. Subsequently, these variables were reviewed by the BCYR research team, which included Aboriginal researchers as well as Aboriginal and non-Aboriginal healthcare professionals and users (mothers) of the BCYR digital assessment. By combining their advice based on their domain knowledge and experience, the final list was narrowed down to 20 variables for analysis, in addition to one output variable (K5) to be predicted. The final selected variable information is listed in Table 1 . Table 1 Selected variables for the prediction model construction. Code Question Feature label Answer options fs1.Q225 I feel lonely like I don’t belong or fit in Feeling Lonely 5: Almost always 4: Often 3: Sometimes 2: A little 1: Hardly ever fs1.Q227 I blame myself when things go wrong Blaming Herself 5: Almost always 4: Often 3: Sometimes 2: A little 1: Hardly ever fs1.Q231 Recently I feel like life is not worth living Life Not Worth Living 1: Never 2: Rarely 3: Sometimes 4: Often 5: Almost always fs1.Q214 I feel strong about being a mum Strong Mum 1: Almost always 2: Often 3: Sometimes 4: A little 5: Hardly ever fs1.Q562 How likely is it that you will do your goals? Goal Likely 1: A lot 2: A fair amount 3: A little bit 4: Not at all fs1.Q904 Managing day to day Managing Day-to-Day 0: Manage well 1: Struggle a bit 2: Struggle a lot fs1.Q534 Client agrees to making a plan to keep safe to deal with the safety worries Keeping Safety Plan 0: Does not agree 1: Client agrees fs1.Q455 Are there ever times when gambling bothers you? Bothered by Gambling 0: Never 1: Sometimes fs1.Q228 I make my family proud Makes Family Proud 1: Almost always 2: Often 3: Sometimes 4: A little 5: Hardly ever fs1.Q454 Are there people close to you gambling? Family Gambles 0: No 1: Sometimes 2: Yes fs1.Q664 How many of these children are in your care? Children in Her Care 0: 0 1: 1 2: 2 3: 3 4: 4 or more fs1.Q450 Have you smoked cigarettes? Smoking Cigarettes in Pregnancy 0: no 1: sometimes 2: yes fs1.Q661 Is this your first pregnancy? First Pregnancy 0: No 1: Yes fs1.Q909 Do you have troubles sleeping? Trouble Sleeping 1: Sleeping well 2: Trouble sleeping (not due to pregnancy/baby) fs1.Q922 Secure housing Need Help with Housing 0: No 1: Yes fs1.Q663 How many previous births have you had? Previous Births 0: 0 1: 1 2: 2 3: 3 4: 4 or more fs1.Q71 Are you feeling worried? Feeling Worried 0: No 1: Yes fs1.Q653 Told partner/husband about pregnancy? Told Partner/Husband 0: No 1: Yes fs1.Q204 Is your male partner angry or controlling? Partner Angry/Controlling 0: No 1: Yes fs1.Q195 Is your male partner moody? Partner Moody 0: No 1: Yes K5 Psychological distress score category low_risk high_risk 3.3. Machine Learning Prediction Model 3.3.1. Traditional ML models Three primary machine learning algorithms—Random Forest (RF), K-Nearest Neighbors (KNN) [ 38 ], and Support Vector Machines (SVM) [ 39 ] —were selected for training and evaluating the prediction model. These choices were made based on their widespread use and proven effectiveness in mental health research, particularly when dealing with tabular data from relatively small datasets [ 40 – 42 ]. RF is a classic ensemble learning technique that builds multiple decision trees during training and combines their decisions to make predictions. This process, while effective, is often seen as opaque because it integrates knowledge learned from multiple individual weak classifiers, making it a black-box model. SVM, another widely used black-box algorithm [ 39 ], aims to find the optimal hyperplane that separates different classes with a maximal margin. Its effectiveness lies in its ability to classify complex, high-dimensional, and non-linear relationships within the data. KNN, in contrast, is an instance-based learning algorithm that classifies a data point based on the classifications of its neighbours [ 38 ]. While KNN may offer some interpretability, considering that it can explain predictions based on the proximity of similar instances using features, his reliance on proximity can be heavily influenced by parameter choices, such as the value of K. Moreover, in high-dimensional spaces, the curse of dimensionality may cause nearest neighbours to lose representativeness, diminishing the effectiveness of proximity-based explanations. Also, KNN's explainability could also fall short of providing a clear understanding of the underlying relationships within the data, which is often crucial information for clinical research. Therefore, there remains a need for further explanation techniques to enhance its interpretability. 3.3.2. Glassbox ML model In addition to the machine learning models mentioned above, we also employed Explainable Boosting Machines (EBM) [ 43 ] for the prediction model. Unlike the previously discussed models, EBM is a glassbox model, which inherently provides explanations for prediction outcomes concerning predictor features, enabling us to interpret and comprehend the model's decisions. Specifically, EBM is a tree-based, gradient boosting Generalized Additive Model (GAM) with automatic pairwise interaction identification [ 43 ]. The contribution of each feature to the final prediction can be visualised and sorted, providing an explanation. The final EBM model is a sum of one-dimensional models for each feature and two-dimensional models for potential pairwise interaction terms. This sum is then passed through a logistic function in the case of classification tasks to ensure the output lies between 0 and 1, representing probabilities of the positive class. In general, EBM can be represented by the following form [ 44 ]: \(\:g\left(E\right[y\left]\right)={\beta\:}_{0}+\sum\:{f}_{i}\left({x}_{i}\right)+\sum\:{f}_{ij}({x}_{i},{x}_{j})\) (1) where \(\:g(\cdot\:)\) is link function, which transforms the expected value of the target variable \(\:y\) depending on the classification or regression problem. \(\:{\beta\:}_{0}\) is the intercept, representing the expected value of the prediction outcome when all the features are equal to zero. \(\:{f}_{i}\left({x}_{i}\right)\:\) is a learned function based on tree-based gradient boosting for each individual feature \(\:{x}_{i}\) , indicating its contribution to the prediction outcome. \(\:{f}_{ij}({x}_{i},{x}_{j})\) is another learned function based on tree-based methods for a pair of features \(\:({x}_{i},{x}_{j})\) , modelling their features’ interactions on the prediction [ 45 ]. To make individual prediction, each function \(\:{f}_{i}\) and \(\:{f}_{ij}\) act as a lookup table per feature and two-dimensional features, respectively, returning a term contribution. These term contributions are then simply added up and passed through the link function \(\:g(\cdot\:)\) for the final prediction. 3.3.3. Prediction performance evaluation To ensure a fair and comparable learning and evaluation process, we employed 10-fold cross-validation. In simpler terms, 10-fold cross-validation involves dividing the designated dataset into ten subsets. The model is then trained on nine of these subsets and validated on the remaining one. This process repeats, cycling through all ten subsets for testing [ 46 ]. In this study, 10-fold cross-validation was utilised for hyperparameter tuning with grid search and performance evaluation. The reported performance of the trained models aggregates the results from 10-fold cross-validation, using the optimal model with the hyperparameter values selected through the model selection procedure. Each model is evaluated using an extensive set of metrics, including accuracy, precision, recall, F1-Score, and Area Under the Curve (AUC), which are common metrics for evaluating classification performance, particularly in health risk assessment prediction tasks. Additionally, we presented the 95% confidence intervals for each performance metric, measuring the uncertainty surrounding them. 3.4. Post-hoc Agnostic Explanation Techniques 3.4.1. SHAP (Shapley Additive exPlanations) SHAP is a model-agnostic technique designed to elucidate the outputs of black-box ML models at both individual and global levels. Rooted in cooperative game theory, SHAP computes Shapley values to determine a fair distribution of the prediction among the features of a given data instance. The Shapley value explanation is represented as an additive feature attribution method using a linear model. The SHAP model can be represented as follows [ 47 ]: $$\:g\left({z}^{{\prime\:}}\right)={\varphi\:}_{0}+\sum\:_{j=1}^{M}{\varphi\:}_{j}{z}_{j}^{{\prime\:}}$$ 2 where \(\:g\) is the model, \(\:{z}^{{\prime\:}}\in\:{\left\{0,\:1\right\}}^{M}\) is the coalition vector, \(\:M\) is the maximum coalition size, and \(\:{\varphi\:}_{j}\in\:\mathcal{\:}\mathcal{R}\) is the feature attribution for feature \(\:j\) , representing Shapley values. The larger the absolute Shapley values, the more important the associated features. Furthermore, Shapley values can be aggregated into global explanations. After running SHAP for every instance, a matrix of Shapley values, with one vector row per data instance and one column vector per feature, can be formulated. Analysing this Shapley value matrix enables us to interpret the entire model at the global level. Kernel SHAP is a specific technique for approximating Shapley values through a sampling process, with the sampled instances weighted using the SHAP kernel based on the coalition’s weight in the Shapley value estimation. 3.4.2. LIME (Local Interpretable Model-Agnostic Explanations) LIME is a widely adopted model-agnostic technique designed to elucidate individual predictions by constructing surrogate local models that approximate the output of a black-box model, thus offering the capacity for local interpretability [ 48 , 49 ]. This technique tests how the machine learning model's predictions change in response to variations in the input data. Specifically, LIME generates a perturbed dataset based on the individual instance of interest and its corresponding predictions from the black-box model. Subsequently, LIME utilizes this perturbed dataset with proximity weightings to train an interpretable model to approximate the black-box model's decision-making process. The LIME local surrogate model optimization problem can be represented as follows. \(\:\underset{g\in\:G}{\text{min}}L\:\left(f,\:g,\:{\pi\:}_{\varvec{x}}\right)+{\Omega\:}\left(g\right)\) (3) The explanation model \(\:g\) for the individual instance \(\:\varvec{x}\) of interest can be a simple interpretable model, e.g., linear regression or a decision tree, that minimize the square loss function \(\:L\) , which measures how close the prediction of the explanation model is to the prediction of the original black-box model \(\:f\) . \(\:{\pi\:}_{\varvec{x}}\) is the proximity measure deciding the perturbed instance neighbourhood around instance \(\:\varvec{x}\) that are used for the explanation. \(\:{\Omega\:}\left(g\right)\) is the regularisation term to control the model complexity. 3.4.3. PDP (Partial Dependence Plots) PDP is a global explanatory, model-agnostic visualisation technique that displays the marginal effect of one or two features on the predicted outcome of a black-box ML model while holding all other features constant [ 50 ]. This technique allows for an intuitive understanding of the relationship between the input features and the model's predictions, especially for non-technical background stakeholders seeking insights into the model's behaviour. Specifically, first, users choose a feature of interest. Once a range of values for the chosen feature is established, based on the user's interest in observing partial dependence, the model computes predictions for each value within that specified range. Simultaneously, all other features are held constant, either fixing them at specific values or by using their average values. Finally, the predicted outcomes against the values of the selected feature are plotted, resulting in a graphical representation of how changes in that specific feature influence the model’s predictions. 4. Results and Discussion 4.1. Prediction Model Performance Table 2 displays the training and testing results for various performance metrics on the BCYR dataset. The best results are highlighted in bold. It is evident that among all the black-box ML models, RF outperformed with an accuracy of 0.829, an F1 score of 0.736, and an AUC of 0.795. Moreover, the glassbox model EBM exhibited the highest performance among all the models, achieving an accuracy of 0.849, an F1 score of 0.771, and an AUC of 0.821, indicating improved predictive capabilities compared to other black-box models. KNN and SVM demonstrated comparable accuracy, slightly below 0.8 (0.798, 0.794), and comparable AUC (0.733, 0.742). In terms of precision and recall metrics, KNN exhibited the highest precision at 0.868 but the lowest recall at 0.514, resulting in an unsatisfactory F1 score of 0.621. Such significant discrepancies in high precision and low recall may potentially miss a considerable number of high-risk cases in the context of mental health risk detection. On the other hand, EBM and RF maintain a commendable balance between precision and recall. Specifically, EBM achieved a precision of 0.829, a recall of 0.727, and an F1 score of 0.771, while RF attained a precision of 0.820, a recall of 0.680, and an F1 score of 0.736. Figure 1 plots the receiver operating characteristic curve (ROC) curves of all the models on one set of testing. It visually shows that the EBM and RF models are the best AUC performers with values of 0.814 and 0.781, respectively, indicating better performance in distinguishing between high-risk and low-risk cases. In summary, EBM and RF outperformed among all the classifiers in achieving high accuracy, F1 score, and AUC, ensuring a reasonable level of both precision and recall for the prediction task, with EBM achieving the best performance. Table 2 Performances of all machine learning models for predicting perinatal mental health risk. Accuracy Mean STD Precision Mean STD Recall Mean STD F1 Mean STD AUC Mean STD RF Training 0.900 (0.8700, 0.9300) 0.05 0.914 (0.8826, 0.9462) 0.05 0.788 (0.7240, 0.8513) 0.10 0.845 (0.7961, 0.8933) 0.08 0.874 (0.8370, 0.9119) 0.06 Testing 0.829 (0.7960, 0.8617) 0.05 0.8200 (0.7524, 0.8869) 0.11 0.680 (0.6099, 0.7501) 0.11 0.736 (0.6859, 0.7851) 0.08 0.795 (0.7581, 0.8318) 0.06 KNN Training 1.000 (1.000, 1.000) 0.00 1.000 (1.000, 1.000) 0.00 1.000 (1.000, 1.000) 0.00 1.000 (1.000, 1.000) 0.00 1.000 (1.000, 1.000) 0.00 Testing 0.798 (0.7524, 0.8430) 0.07 0.868 (0.7860, 0.9491) 0.13 0.514 (0.3846, 0.6426) 0.21 0.621 (0.5204, 0.7216) 0.16 0.733 (0.6709, 0.7948) 0.10 EBM Training 0.886 (0.8797, 0.8927) 0.01 0.899 (0.8882, 0.9091) 0.02 0.764 (0.7505, 0.7770) 0.02 0.826 (0.8152, 0.8359) 0.02 0.858 (0.8506, 0.8661) 0.01 Testing 0.849 (0.8170, 0.8814) 0.05 0.829 (0.7689, 0.8900) 0.10 0.727 (0.6599, 0.7946) 0.11 0.771 (0.7169, 0.8245) 0.09 0.821 (0.7829, 0.8593) 0.06 SVM Training 0.858 (0.8275, 0.8894) 0.05 0.920 (0.8943, 0.9466) 0.04 0.652 (0.5804, 0.7236) 0.12 0.760 (0.7055, 0.8150) 0.09 0.812 (0.7715, 0.8517) 0.06 Testing 0.794 (0.7514, 0.8373) 0.07 0.805 (0.7195, 0.8906) 0.14 0.570 (0.4604, 0.6796) 0.18 0.650 (0.5675, 0.7335) 0.13 0.742 (0.6880, 0.7966) 0.09 4.2. Explanation Results 4.2.1. Global interpretation by EBM and SHAP As EBM demonstrates high predictive performance and operates as a transparent glassbox model, we generated the explanation from EBM and illustrated the global feature importance over the whole data set in Fig. 2 . Longer bars in the figure indicate higher importance of features in the model's predictions. It is noteworthy that specific features exert a more significant influence on the decision-making process of the EBM model. For instance, the feature “Feeling Lonely” emerges as the most impactful, suggesting that a participant’s response to a reflection concerning feelings of loneliness, might serve as a robust predictor for perinatal mental health risk. This is followed by “Blaming Herself” and “Makes Family Proud,” indicating that these two features have a higher contribution to the overall prediction (risk or protective). Other features, such as “Managing Day-to-Day” and “Strong Mum,” also demonstrate a notable impact in relation to the K5 target outcome, which is aligned with previous study findings [ 9 ]. Moreover, the figure shows specific multi-dimensional interactions that add increased weighting to the model’s predictions. For example, the feature “Feeling Lonely” in combination with features such as “Life Not Worth Living,” “Blaming Herself,” or “Partner Angry/Controlling” places greater predictive power. Likewise, the joint effects of “Blaming Herself” and “Family Gambles” or “Makes Family Proud” are also identified as important feature combinations affecting the model predictions. For the high-performing black-box RF model, we utilised SHAP to also gain insights into the model’s explanation. We displayed the global feature importance ranking in Fig. 3 . The most important features, including “Feeling Lonely,” “Blaming Herself,” “Managing Day-to-Day,” “Life Not Worth Living,” and “Makes Family Proud,” exactly overlap with the top features selected from EBM. Additionally, features related to questions concerning “Family Gambles,” “Strong Mum,” “Goal Likely,” and “Trouble Sleeping” were subsequently ranked, aligning with EBM’s ranking tier as well. 4.2.2. Local interpretation by EBM, SHAP, LIME EBM can also provide an interpretation for the model’s prediction on an individual instance. Figure 4 shows one Aboriginal woman who is low risk, labelled “Instance I,” it shows the specific contributions of each feature and their interactions in combination toward risk prediction for this individual by EBM. EBM accurately predicts this individual as “low risk” (class = 0), with a high probability score of 0.904. The y-axes of the plot show impactful features and their corresponding values in brackets. Specifically, contributing features highlighted in blue, such as “Blaming Herself” = 1.00 (hardly ever) and “Feeling Lonely” = 1.00 (hardly ever), contribute to shifting the model's prediction towards the low-risk category. These two features stand out, suggesting that hardly ever feeling lonely and hardly ever blaming herself are strongly protective factors for this Aboriginal mother. Additionally, the combination of “Blaming Herself” = 1.00 (hardly ever) and “Family Gambles” = 0.00 (loved ones do not gamble) shows an additive positive influence, pushing the prediction towards low risk. Conversely, contributing features displayed in orange in the figure, would have pushed the model's prediction away from low risk. For example, “Children in Her Care” = 3 (having three children in care) and “Strong Mum” = 3 (sometimes), and “Makes Family Proud” = 2 (often) and “Partner Moody” = 1 (yes), are risk factors that increase this woman’s risk. We noted that “Makes Family Proud” and “Strong Mum” were highly protective only when women selected the top rating: ‘Always’. Looking at the SHAP techniques, we generated force plots in Fig. 5 to visualise the contributions of important features to the same (“Instance I”) woman’s prediction. The lower prediction score of 0.12 indicates that this participant is at low risk. The key features influencing the prediction for this individual are depicted in red and blue. Red indicates features that elevated the model's score towards high risk. Blue signifies features that reduced the risk. Features having greater impact on prediction scores are located closer to the dividing boundary between red and blue in the figure. Therefore, the strongly protective factors contributing to this low-risk prediction, including “hardly ever blames herself,” “hardly ever feels lonely,” and “close loved ones do not gamble,” align with the EBM’s individual interpretation. Additionally, “sometimes feels strong about being a mum” and “moody partner” tended to push the prediction towards higher risk, which again is similar to EBM’s interpretation. We employed an additional technology, the LIME technique, to further explore individual influencing features for the same woman. Overall, the results were very similar to EBM and SHAP technologies. Figure 6 illustrates that the most influential features in predicting low risk were “hardly ever feels lonely,” “hardly ever blames herself,” “never feels life is not worth living,” and “close loved ones do not gamble.” Interestingly LIME uniquely identified that “needs no help with housing” was a contributor to low risk, not identified by EBM and SHAP. Figure 7 provides the EBM’s local explanation for another Aboriginal woman labelled “Instance II,” who was at greater risk. It was accurately predicted with a high risk of perinatal mental health problems and a probability score of 0.655. The chart reveals that the factors significantly influencing the prediction decision towards high risk include “Feeling Lonely”=2 (feels a little lonely), “Trouble Sleeping”=2 (having trouble sleeping), “Family Gambles”=2 (close loved ones gamble), “Makes Family Proud”=3 (sometimes makes her family proud), and “Goal Likely”=2 (a fair amount). Meanwhile, features in blue, such as “Blaming Herself”=1 (hardly ever blames herself), “Managing Day-to-Day”=0 (manages day-to-day well), “Life Not Worth Living”=1 (never feels life is not worth living), “Strong Mum”=1 (almost always feels strong about being a mum), “Children in Her Care”=1 (having one child in care), and “Previous Births”=1 (one previous birth), serve as protective factors in this specific instance. Figure 8 shows the prediction interpretation for the same individual using SHAP. A consistent group of answers to the questions that drive the prediction towards a high risk was identified, including “Family Gambles,” “Feeling Lonely,” “Makes Family Proud,” “Goal Likely,” and “Trouble Sleeping,” although the significance level ranking shows slight differences. Meanwhile, not “Blaming Herself” and “Managing Day-to-Day” were identified as mitigating factors that attempt to reduce this high-risk prediction. Figure 9 provides LIME’s local interpretation of the same individual. It selected the same group of risk factors as SHAP, except for “Trouble Sleeping,” which was not chosen by LIME but was selected by both SHAP and EBM. The protective factors also largely align with the other two methods, except for “Need Help With Housing” =0, chosen as a protective factor by LIME, which was not picked as a top protective factor by the other two methods. 4.2.3. High influential factor interpretation by PDP Through interpretation conducted using both glassbox and model-agnostic techniques, we identified two highly significant questions impacting the prediction outcomes: 'Blaming Herself' and 'Feeling Lonely.' We further adopted their corresponding PDP plots to reveal their individual relationships with the average predicted outcome shown in Fig. 10 . In the case of 'Blaming Herself,' mothers who hardly ever blame themselves are associated with a low risk. However, starting from 'A Little' and progressing to 'Sometimes,' 'Often,' and 'Almost Always,' there is a significant increase in the predicted risk of perinatal mental health issues. This suggests that as the frequency of self-blame rises, even starting from 'A Little,' the associated risk shows a significant increasing trend. There is not much difference in the impact between the categories 'Sometimes,' 'Often,' and 'Almost Always,' as they all show a significant level of relation with high risk. Similar observations were made for the case of 'Feeling Lonely,' where mothers who reported 'Hardly Ever' feeling lonely are associated with a low risk. We further generated a PDP plot to visualise the interplay between these two significant factors. The heatmap colour gradient represents the predicted risk level, with light cream indicating a low risk (closer to 0) and purple representing a high risk (closer to 1). Mothers who reported “Hardly Ever” blaming themselves and “Hardly Ever” feeling lonely are in the lightest zone, suggesting the lowest predicted risk. However, the deeper colour zone is pronounced for respondents who reported feeling lonely starting from “A Little” and beyond and blaming themselves starting from “A Little” and beyond. This combined emotional state of frequent loneliness and self-blaming puts them at a much higher predicted risk for perinatal mental health issues. Regardless of whether the response was “Sometimes,” “Often,” or “Always” for loneliness and self-blame, it led to the highest level of risk, with almost the same effect. 5. CONCLUSIONs AND FINAL REMARKS In conclusion, our results reveal that both the glassbox EBM model and the blackbox RF model, which are ensemble-based methods, excel in predicting perinatal mental health issues among Aboriginal mothers. The EBM achieved an accuracy of 0.849 and an AUC of 0.821, while the random forest model achieved an accuracy of 0.829 and an AUC of 0.795. Additionally, we compared the explainability generated by EBM and post-hoc agnostic XAI technologies including SHAP and LIME at both the global and individual levels. Generally, both EBM and SHAP provide global explanations that highlight questions about “Feeling Lonely” and “Blaming Herself” as dominant factors influencing the overall prediction outcome. Additionally, items regarding “Makes Family Proud,” “Life Not Worth Living,” “Managing Day-to-Day,” “Strong Mum,” and “Family Gambles” are also identified as top-ranking questions. These show minimal variations across different XAI interpretation methods. At the individual level, we examined two cases—one high risk and one low risk—as examples, using EBM, SHAP, and LIME to generate visual maps. These maps clearly displayed significant protective and risk factors contributing to the prediction outputs, with very minor variations in the different XAI interpretation methods. Across all three XAI interpretation methods, emotional states related to feelings of loneliness, self-blame, and perceptions of self-worth consistently emerge as powerful predictors for perinatal mental health issues for Aboriginal mothers. Additionally, certain features in combination such as 'Feeling Lonely' and 'Life Not Worth Living,' also influenced predictions of perinatal maternal risk. By analysing the combination of global and local insights gained from these XAI methods, we can achieve a better and more transparent understanding of the data-driven model's decision-making process regarding Aboriginal mothers' perinatal mental health and associated factors, and even delve into individual-level insights. We anticipate that this XAI-assisted and co-designed prediction model can enhance clinicians' and health services' capacity and confidence to provide culturally sensitive clinical reasoning and accurate interpretation of screening results that reduces existing biases. We predict it may effectively facilitate early and accurate identification of perinatal mental-health issues, enhance relevant and quality supports and referrals, strengthen culturally secure care and minimise unnecessary child protection notifications. Through combining transparent XAI outputs with lived experience wisdom this clinical prediction model may provide new insights, knowledge, and mechanisms for addressing perinatal mental-health issues among Aboriginal women, facilitate closer multidisciplinary collaboration, improve communication, and build trust between new techniques, researchers, clinicians, and Aboriginal community. This would result in continuous improvement in service quality and efficiency, cost saves and support Aboriginal families to flourish and thrive. There are limitations that need to be addressed in future work. Firstly, the sample size used for building the prediction model is limited, despite the high number of features in the de-identified data. As more data is collected through the BCYR program across different regions, future opportunities will arise to increase the sample size, enhancing the model's performance. Secondly, the current visual maps generated from the XAI techniques will be improved through the co-design process to enhance readability and explainability, especially for Aboriginal women and clinicians. This user-friendly and culturally sensitive visual prediction model will ensure accurate and efficient interpretation by all stakeholders involved. Abbreviations (ACC) Accuracy (AD) Dementia (AI) Artificial Intelligence (AUC) Area Under the Curve (AUROC) Area Under the Receiver Operating Characteristic Curve (BCYR) “Baby Coming You Ready?” (CMD) Cardiometabolic Diseases (CNN) Convolutional Neural Network (DNN) Deep Neuronal Network (EBM) Explainable Boosting Machine (HER/ EMR) Electronic Health Record or Electronic Medical Record (IEA) Image-emotion Alignment (ITA) Image-text Alignment (LIME) Local Interpretable Model-Agnostic Explanations (LR) Logistic Regression (ML) Machine Learning (NA) No Answer (NB) Naïve Bayes (NLP) Natural Language Processing (PDP) Partial Dependence Plots (PHC) Primary Health Care (PNDA) Perinatal depression (PRECIS) Precision (RF) Random Forest (ROC) Receiver Operating Characteristic (SHAP) SHapley Additive exPlanations (XAI) Explainable Artificial Learning Declarations Ethics approval and consent to participate Ethics approval was obtained from the Human Research Ethics Committee (HREC) - Western Australia Research Governance Service: RGS000000649; Murdoch University: 2021/101; Western Australian Aboriginal Health Ethics Committee (WAAHEC): HREC553. Access to deidentified data for this study made available only from consenting participants. A consent button to either use or not to use deidentified data for research purposes is embedded in the digital application. Consent for publication Not applicable. Competing Interests Not applicable. Funding The work was supported by the Western Australian Future Health Research and Innovation Fund (Grant ID IC2023-GAIA/18), and J. Kotz and G. Wang are supported by the Google Inclusion Research Award. Author Contribution G.W., W.K., R.M., R.W., and J.K. designed the work; G.W., W.K., H.B., and J.Q. performed the experiments and data analysis; G.W., J.Q., and J.K. drafted the manuscript; G.W. and J.K. substantively revised the manuscript. All authors interpreted the data and results and reviewed the manuscript. Acknowledgements Not Applicable. Availability of data and materials The data supporting the findings of this study are not openly available. Requests for more detailed information can be directed to the corresponding author. References Brummelte S, Galea LA. Postpartum depression: Etiology, treatment and consequences for maternal care. Horm Behav. 2016;77:153–66. Mitchell AR, Gordon H, Lindquist A, Walker SP, Homer CS, Middleton A, Cluver CA, Tong S, Hastie R. Prevalence of perinatal depression in low-and middle-income countries: A systematic review and meta-analysis. JAMA Psychiatry. 2023;80(5):425. Hummel AD, Ronen K, Bhat A, Wandika B, Choo EM, Osborn L, Batra M, Kinuthia J, Kumar M, Unger JA. Perinatal depression and its impact on infant outcomes and maternal-nurse SMS communication in a cohort of Kenyan women. BMC Pregnancy Childbirth. 2022;22(1):723. De Backer K, Wilson CA, Dolman C, Vowles Z, Easter A. Rising rates of perinatal suicide. 2023, 381:e075414. Australian Institute of Health and Welfare. Child Protection Australia 2019-20. Child welfare series no. 74. Cat. no. CWS 78. Australian Institute of Health and Welfare; 2021. Indigenous health and wellbeing. [ https://www.aihw.gov.au/reports/australias-health/indigenous-health-and-wellbeing] . Owais S, Faltyn M, Johnson AV, Gabel C, Downey B, Kates N, Van Lieshout RJ. The perinatal mental health of indigenous women: A systematic review and meta-analysis. Can J Psychiatry. 2020;65(3):149–63. Lima F, Shepherd C, Wong J, O’Donnell M, Marriott R. Trends in mental health related contacts among mothers of Aboriginal children in Western Australia (1990–2013): A linked data population-based cohort study of over 40 000 children. BMJ Open. 2019;9(7):e027733. Ratajczak T. Exploring Cultural and Clinical Factors Contributing to Aboriginal and Torres Strait Islander Women’s Resilience: A Study Utilising the Baby Coming You Ready Program. WA: Murdoch University; 2024. Kotz J, Bennett E, Reibel T, AhChee V, Marriott R. Baby coming you ready' an innovative screening and assessment tool for perinatal mental health with Aboriginal families: A protocol for pilot evaluation. Australian J Child Family Health Nurs. 2021;18(2):19–26. Gee G, Dudgeon P, Schultz C, Hart A, Kelly K. Aboriginal and Torres Strait Islander social and emotional wellbeing. In: Working together: Aboriginal and Torres Strait Islander mental health and wellbeing principles and practice. Edited by Dudgeon P, Milroy H, Walker R, vol. 2; 2014: 55–68. Novick AM, Kwitowski M, Dempsey J, Cooke DL, Dempsey AG. Technology-based approaches for supporting perinatal mental health. Curr Psychiatry Rep. 2022;24(9):419–29. Bilal AM, Fransson E, Bränn E, Eriksson A, Zhong M, Gidén K, Elofsson U, Axfors C, Skalkidou A, Papadopoulos FC. Predicting perinatal health outcomes using smartphone-based digital phenotyping and machine learning in a prospective Swedish cohort (Mom2B): Study protocol. BMJ Open. 2022;12(4):e059033. Cellini P, Pigoni A, Delvecchio G, Moltrasio C, Brambilla P. Machine learning in the prediction of postpartum depression: A review. J Affect Disord. 2022;309:350–7. Wang S, Pathak J, Zhang Y. Using electronic health records and machine learning to predict postpartum depression. MEDINFO 2019: Health and Wellbeing e-Networks for All. IOS; 2019. pp. 888–92. Zhang Y, Wang S, Hermann A, Joly R, Pathak J. Development and validation of a machine learning algorithm for predicting the risk of postpartum depression among pregnant women. J Affect Disord. 2021;279:1–8. Zhong M, Van Zoest V, Bilal AM, Papadopoulos F, Castellano G. Unimodal vs. multimodal prediction of antenatal depression from smartphone-based survey data in a longitudinal study. In: 2022 International Conference on Multimodal Interaction: 2022; New York ; 2022: 455–467. Javed F, Gilani SO, Latif S, Waris A, Jamil M, Waqas A. Predicting risk of antenatal depression and anxiety using multi-layer perceptrons and support vector machines. J Personalized Med. 2021;11(3):199. Garriga R, Mas J, Abraha S, Nolan J, Harrison O, Tadros G, Matic A. Machine learning model to predict mental health crises from electronic health records. Nat Med. 2022;28(6):1240–8. Kelly B. Culturally Safe Mental Health Screening for Aboriginal Mothers: An exploration of the value of Baby Coming You Ready program in child health settings. WA: Murdoch University; 2024. Habib A-R, Crossland G, Patel H, Wong E, Kong K, Gunasekera H, Richards B, Caffery L, Perry C, Sacks R. An artificial intelligence computer-vision algorithm to triage otoscopic images from Australian Aboriginal and Torres Strait Islander children. Otology Neurotology. 2022;43(4):481–8. Scheetz J, Koca D, McGuinness M, Holloway E, Tan Z, Zhu Z, O’day R, Sandhu S, MacIsaac RJ, Gilfillan C. Real-world artificial intelligence-based opportunistic screening for diabetic retinopathy in endocrinology and indigenous healthcare settings in Australia. Sci Rep. 2021;11(1):15808. Chando S, Craig JC, Burgess L, Sherriff S, Purcell A, Gunasekera H, Banks S, Smith N, Banks E, Woolfenden S. Developmental risk among Aboriginal children living in urban areas in Australia: The Study of Environment on Aboriginal Resilience and Child Health (SEARCH). BMC Pediatr. 2020;20:1–13. Jeong K, Mallard AR, Coombe L, Ward J. Artificial intelligence and prediction of cardiometabolic disease: Systematic review of model performance and potential benefits in indigenous populations. Artif Intell Med. 2023;139:102534. Nagendran M, Chen Y, Lovejoy CA, Gordon AC, Komorowski M, Harvey H, Topol EJ, Ioannidis JPA, Collins GS, Maruthappu M. Artificial intelligence versus clinicians: Systematic review of design, reporting standards, and claims of deep learning studies in medical imaging. BMJ (Online) 2020, 368. Letzgus S, Müller K-R. Towards transparent and robust data-driven wind turbine power curve models. arXiv preprint 2023:arXiv:230409835. Holzinger A, Saranti A, Molnar C, Biecek P, Samek W. Explainable AI methods - A brief overview. In: xxAI - Beyond Explainable AI: International Workshop, Held in Conjunction with ICML 2020, July 18, 2020, Vienna, Austria, Revised and Extended Papers. Edited by Holzinger A, Goebel R, Fong R, Moon T, Müller K-R, Samek W. Cham: Springer International Publishing; 2022: 13–38. Byeon H. Advances in machine learning and explainable artificial intelligence for depression prediction. Int J Adv Comput Sci Appl 2023, 14(6). Loh HW, Ooi CP, Seoni S, Barua PD, Molinari F, Acharya UR. Application of explainable artificial intelligence for healthcare: A systematic review of the last decade (2011–2022). Computer Methods and Programs in Biomedicine 2022, 226:107161. Jothi N, Husain W. Predicting generalized anxiety disorder among women using Shapley value. J Infect Public Health. 2021;14(1):103–8. Ikram M, Shaikh NF, Vishwanatha JK, Sambamoorthi U. Leading predictors of covid-19-related poor mental health in adult Asian Indians: An application of extreme gradient boosting and Shapley additive explanations. Int J Environ Res Public Health. 2022;20(1):775. Hussna AU, Trisha II, Ritun IJ, Alam MGR. Covid-19 impact on students' mental health: Explainable AI and classifiers. In: International Conference on Decision Aid Sciences and Application (DASA): 2021; Sakheer, Bahrain : IEEE; 2021: 847–851. Adarsh V, Gangadharan GR. Applying explainable artificial intelligence models for understanding depression among IT workers. IT Prof. 2022;24(5):25–9. Joyce DW, Kormilitzin A, Smith KA, Cipriani A. Explainable artificial intelligence for mental health through transparency and interpretability for understandability. npj Digit Med. 2023;6(1):6. Kotz J. Kalyakool Moort-Always Family: Strong culture, strong care, strong families: Codesigning a culturally considered approach to perinatal screening and support for the first 1000 days of parenthood from conception to infancy. Murdoch University; 2021. Brinckley M-M, Calabria B, Walker J, Thurber KA, Lovett R. Reliability, validity, and clinical utility of a culturally modified Kessler scale (MK-K5) in the Aboriginal and Torres Strait Islander population. BMC Public Health. 2021;21(1):1111. Kwok SWH, Kotz J, Reibel T, Wang G, Walker R, Marriott R. Coupling machine learning models with an innovative technology-based screening tool for identifying psychological distress among Aboriginal perinatal mothers. In: 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) : 2023; Sydney: IEEE; 2023: 1–4. Kramer O. K-nearest neighbors. In: Dimensionality reduction with unsupervised nearest neighbors. Edited by Kramer O. Berlin, Heidelberg: Springer Berlin Heidelberg; 2013: 13–23. Suthaharan S. Support vector machine. In: Machine learning models and algorithms for big data classification: Thinking with examples for effective learning. Edited by Suthaharan S. Boston, MA: Springer US; 2016: 207–235. Mohamed ES, Naqishbandi TA, Bukhari SAC, Rauf I, Sawrikar V, Hussain A. A hybrid mental health prediction model using support vector machine, multilayer perceptron, and random forest algorithms. Healthc Analytics. 2023;3:100185. Cho G, Yim J, Choi Y, Ko J, Lee S-H. Review of machine learning algorithms for diagnosing mental illness. Psychiatry Invest. 2019;16(4):262. Khondoker M, Dobson R, Skirrow C, Simmons A, Stahl D. A comparison of machine learning methods for classification using simulation with multiple real data examples from mental health studies. Stat Methods Med Res. 2016;25(5):1804–23. Nori H, Jenkins S, Koch P, Caruana R. Interpretml: A unified framework for machine learning interpretability. arXiv preprint 2019:arXiv:1909.09223. Lou Y, Caruana R, Gehrke J, Hooker G. Accurate intelligible models with pairwise interactions. In: 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining : 2013; 2013: 623–631. Thimoteo LM, Vellasco MM, Amaral J, Figueiredo K, Yokoyama CL, Marques E. Explainable artificial intelligence for COVID-19 diagnosis through blood test variables. J Control Autom Electr Syst. 2022;33(2):625–44. Poldrack RA, Huckins G, Varoquaux G. Establishment of best practices for evidence for prediction: A review. JAMA Psychiatry. 2020;77(5):534–40. Lundberg SM, Erion GG, Lee S-I. Consistent individualized feature attribution for tree ensembles. arXiv preprint 2018:arXiv:1802.03888. Ribeiro MT, Singh S, Guestrin C. Why should I trust you? Explaining the predictions of any classifier. In: 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco, California, USA: Association for Computing Machinery; 2016: 1135–1144. Hong W, Lu Y, Zhou X, Jin S, Pan J, Lin Q, Yang S, Basharat Z, Zippi M, Goyal H. Usefulness of random forest algorithm in predicting severe acute pancreatitis. Front Cell Infect Microbiol. 2022;12:893294. Belle V, Papantonis I. Principles and practice of explainable machine learning. Front Big Data. 2021;4:688969. Footnotes Government policies ensured Aboriginal children were systematically and forcibly removed from their families, communities and culture from the mid-1800s for almost a century. These children are referred to as the Stolen Generation or Stolen Children. These children were denied access to their culture and were punished if they spoke their language. The impacts of this are still being felt today (Healing Foundation, 2022). Glassbox models allow users to trace the decision-making process back to individual data points or variables, making it easier to understand and interpret an algorithm's output. Post-hoc agnostic models allow the decision-making mechanism to be a black box, and then develops a model that generates explanations after a recommendation has been made. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4687233","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":335065109,"identity":"6d9e2382-f211-43f7-8566-9eb018007803","order_by":0,"name":"Guanjin Wang","email":"data:image/png;base64,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","orcid":"","institution":"Murdoch University","correspondingAuthor":true,"prefix":"","firstName":"Guanjin","middleName":"","lastName":"Wang","suffix":""},{"id":335065111,"identity":"4fd5f088-05a2-4dee-a3dc-d70b5f144d65","order_by":1,"name":"Hachem Bennamoun","email":"","orcid":"","institution":"Murdoch University","correspondingAuthor":false,"prefix":"","firstName":"Hachem","middleName":"","lastName":"Bennamoun","suffix":""},{"id":335065112,"identity":"60739a74-5c6d-41df-9a16-b6fb2271399a","order_by":2,"name":"Wai Hang Kwok","email":"","orcid":"","institution":"Edith Cowan University","correspondingAuthor":false,"prefix":"","firstName":"Wai","middleName":"Hang","lastName":"Kwok","suffix":""},{"id":335065113,"identity":"b2fc9112-fb9f-45a2-ac69-d124ec990d97","order_by":3,"name":"Jenny Paola Ortega Quimbayo","email":"","orcid":"","institution":"Murdoch University","correspondingAuthor":false,"prefix":"","firstName":"Jenny","middleName":"Paola Ortega","lastName":"Quimbayo","suffix":""},{"id":335065114,"identity":"d3274f26-6018-474b-b591-09776a87368b","order_by":4,"name":"Bridgette Kelly","email":"","orcid":"","institution":"Murdoch University","correspondingAuthor":false,"prefix":"","firstName":"Bridgette","middleName":"","lastName":"Kelly","suffix":""},{"id":335065115,"identity":"2839aab6-3fae-4ace-b77a-a183de378453","order_by":5,"name":"Trish Ratajczak","email":"","orcid":"","institution":"Murdoch University","correspondingAuthor":false,"prefix":"","firstName":"Trish","middleName":"","lastName":"Ratajczak","suffix":""},{"id":335065116,"identity":"471faacd-4e76-4cd1-b4a1-3c2f0f452a27","order_by":6,"name":"Rhonda Marriott","email":"","orcid":"","institution":"Murdoch University","correspondingAuthor":false,"prefix":"","firstName":"Rhonda","middleName":"","lastName":"Marriott","suffix":""},{"id":335065117,"identity":"27bd64f7-5fc2-496a-9725-8e58a07f6499","order_by":7,"name":"Roz Walker","email":"","orcid":"","institution":"Murdoch University","correspondingAuthor":false,"prefix":"","firstName":"Roz","middleName":"","lastName":"Walker","suffix":""},{"id":335065118,"identity":"08d933a2-bf25-441a-a269-8852c7b7196c","order_by":8,"name":"Jayne Kotz","email":"","orcid":"","institution":"Murdoch University","correspondingAuthor":false,"prefix":"","firstName":"Jayne","middleName":"","lastName":"Kotz","suffix":""}],"badges":[],"createdAt":"2024-07-04 14:25:42","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4687233/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4687233/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":61662383,"identity":"6f3f7c36-4e2a-46bc-8b92-4a24ae0fa1fd","added_by":"auto","created_at":"2024-08-02 15:09:30","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":45944,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver Operating Characteristic curve (ROC) plot of all machine learning models predicting perinatal mental health risk.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/2fe994116c0e88089a82ef39.png"},{"id":61663102,"identity":"466acb64-b149-4da8-ae84-e6fe2fcf5b43","added_by":"auto","created_at":"2024-08-02 15:17:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":14612,"visible":true,"origin":"","legend":"\u003cp\u003eGlobal feature importance interpretation from glassbox model EBM.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/bd9c68e33ab366ad5087edde.png"},{"id":61663818,"identity":"18d51777-6a5e-4c83-bdde-b7fd141b61a6","added_by":"auto","created_at":"2024-08-02 15:25:30","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":44426,"visible":true,"origin":"","legend":"\u003cp\u003eGlobal feature importance interpretation from post-hoc agnostic SHAP.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/8f1a4be2803a22a9fa990b92.png"},{"id":61663101,"identity":"00553042-8611-461b-ab37-a76d164ba068","added_by":"auto","created_at":"2024-08-02 15:17:30","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":48271,"visible":true,"origin":"","legend":"\u003cp\u003eLocal explanation for Instance I by EBM\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/776d649a5f1d1e1ffd0f2c27.png"},{"id":61662392,"identity":"8ee16d12-3f1b-4c76-8455-1633fbc09990","added_by":"auto","created_at":"2024-08-02 15:09:30","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":33029,"visible":true,"origin":"","legend":"\u003cp\u003eLocal explanation for Instance I by SHAP\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/bd6e0304bbef2ed077a3980e.png"},{"id":61664128,"identity":"68ba19ca-1d2f-45eb-8a75-5f088f275ffe","added_by":"auto","created_at":"2024-08-02 15:33:30","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":42754,"visible":true,"origin":"","legend":"\u003cp\u003eLocal explanation for Instance I by LIME.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/61eead2a8804cffe59f367e5.png"},{"id":61663106,"identity":"96ef4f45-56cb-4918-8ad5-293c1319212f","added_by":"auto","created_at":"2024-08-02 15:17:30","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":49088,"visible":true,"origin":"","legend":"\u003cp\u003eLocal explanation for Instance II by EBM\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/e725cf96c210e572a09e98f6.png"},{"id":61662389,"identity":"cf9ac532-de16-4b8e-8bdb-852394673ec6","added_by":"auto","created_at":"2024-08-02 15:09:30","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":34183,"visible":true,"origin":"","legend":"\u003cp\u003eLocal explanation for Instance II by SHAP\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/2714ab1fca08f8d66db4256e.png"},{"id":61663103,"identity":"0b207186-078a-4fc3-ac55-00b0621cf56e","added_by":"auto","created_at":"2024-08-02 15:17:30","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":54758,"visible":true,"origin":"","legend":"\u003cp\u003eLocal explanation for Instance II by LIME.\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/bb4689cfe6339e85504a2850.png"},{"id":61662386,"identity":"540ccd8f-a2da-429c-84bd-f67d63d9127d","added_by":"auto","created_at":"2024-08-02 15:09:30","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":33383,"visible":true,"origin":"","legend":"\u003cp\u003ePDP for Blaming Herself; PDP for Feeling Lonely, PDP for Blaming Herself vs Feeling Lonely\u003c/p\u003e","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/68b3622fa75c27c9e6546cea.png"},{"id":64578414,"identity":"490fbb1e-7262-4d3a-a33e-af11174d6eff","added_by":"auto","created_at":"2024-09-16 04:56:42","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1295554,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4687233/v1/fa858fc8-788c-4f34-b5cc-4792bce44b16.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Decoding Perinatal Mental Health: Investigating Protective and Risk Factors and Predictive Insights for Aboriginal Perinatal Mental Health through Explainable Machine Learning","fulltext":[{"header":"1. Background","content":"\u003cp\u003ePerinatal depression and anxiety negatively impact the health and well-being of mothers and babies, and disrupts maternal /infant bonding [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Recent studies highlighted the significant association between perinatal depression and adverse outcomes, including suicidal behaviours and self-harm thoughts during and after pregnancy. AR Mitchell, H Gordon, A Lindquist, SP Walker, CS Homer, A Middleton, CA Cluver, S Tong and R Hastie [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] emphasised the increased risk of preterm birth, stillbirth, and suicide associated with perinatal depression. Furthermore, AD Hummel, K Ronen, A Bhat, B Wandika, EM Choo, L Osborn, M Batra, J Kinuthia, M Kumar and JA Unger [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] highlighted the association of perinatal depression with adverse infant outcomes such as preterm birth, intrauterine growth restriction, and low birth weight. The loss of an infant's mother through suicide profoundly impacts the infant's social and emotional well-being [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Many Aboriginal women experience strong social and emotional well-being and have flourishing infants and families. However, at a population level, too many Aboriginal women face the increased risk of triggering or worsening depression/anxiety directly resulting from the enduring challenges, barriers and adversities from colonisation, cultural disruption and past and present policies such as the Stolen Generations\u003ca class=\"FNLink\" href=\"#Fn1\" id=\"#FNLinkFn1\"\u003e\u003c/a\u003e. These include poverty, racism, intergenerational and complex trauma, racism, cultural bias, loss of cultural identity and other inequities [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. A systematic review by S Owais, M Faltyn, AV Johnson, C Gabel, B Downey, N Kates and RJ Van Lieshout [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] indicated Aboriginal women face a 38% higher chance of experiencing depression, are 79% more susceptible to mental-health problems during pregnancy, and 30% more likely to endure mental health complexities post giving birth. A Western Australian study revealed that between 1997 to 2013, one in three Aboriginal babies were born to mothers who sought hospital care for mental-health issues related to substance abuse, depression and/or anxiety [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Despite routine screening for perinatal depression and anxiety in Australia for over 20 years, the gap in Aboriginal mothers\u0026rsquo; and infants\u0026rsquo; health and wellbeing remains unacceptable across all key indicators. This is evident in disproportionately higher rates of premature births, low-birth weight babies and child removal [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Conventional health systems approach to antenatal/postnatal care and screening are often culturally insensitive and retraumatising for Aboriginal women [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Risk-focused perinatal screens and assessments with Aboriginal women frequently exacerbate feelings of alienation and disengagement from potentially supportive care [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. There is an urgent need for, culturally safe and effective trauma-aware and healing informed screening for social, emotional, mental-health and well-being that includes relevant supportive and strength-based follow-up care for Aboriginal mothers.\u003c/p\u003e \u003cp\u003eThe \u003cem\u003eBaby Coming You Ready (BCYR)\u003c/em\u003e program [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], was co-designed to overcome these barriers and challenges faced by Aboriginal parents during their perinatal care. The BCYR Program centres around a digitised, strengths-based, culturally safe, and holistic perinatal assessment that incorporates all seven elements of Aboriginal peoples\u0026rsquo; social and emotional well-being [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. The assessment uses iPads with touchscreen images depicting common strengths, worries and past and present occurrences. Aboriginal voice-overs on each slide guide the reflective engagement between the mother and her midwife or health professional. Mothers choose images they relate to while engaging in self-reflection, creating their own personalised story, then prioritise their strengths and concerns before designing their own solutions. Currently the BCYR program is operationalised as a model of care in all six pilot sites in Western Australia (WA), and effectively replaces all current required screens for mental-health, family and domestic violence (FDV), tobacco, alcohol and other drugs (AOD). Whilst the successful pilot demonstrated increased trust, engagement, honest disclosure and self-directed management plans, it found that some midwives and managers lacked confidence in culturally considered assessments [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. They continued to overly rely on the embedded risk-screen score (Kessler-5 distress scale) which undermines the trauma-aware and healing-informed approach. It weakens the potential of the BCYR assessment and health clinicians to effectively support Aboriginal mothers\u0026rsquo; strengths and self-determined solutions, and ultimately contribute to improve maternal/infant outcomes and public health cost savings.\u003c/p\u003e \u003cp\u003eOver the last decade, with the advancements in digital health and computational capabilities, numerous studies have focused on technology-based approaches to advance support for perinatal mental health [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. This includes the development of Artificial Intelligence (AI)-based models, particularly machine learning and deep learning-based models, for predicting perinatal mental-health conditions [\u003cspan additionalcitationids=\"CR15 CR16 CR17\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. These AI-assisted decision-support approaches in mainstream populations have potential to facilitate early and accurate detection of depression, thereby enabling timely interventions. They may also assist in identifying patterns that might be overlooked in manual assessments, leading to improved clinical efficiency and outcomes [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. To mitigate the identified service gap in Aboriginal-focused perinatal mental health screening and care, our aim is to integrate Aboriginal ways of being and knowing with AI technology to develop a culturally considered, strengths-focused clinical prediction model. Specifically, our approach introduces eXplainable Artificial Intelligence (XAI) technologies, including glassbox models\u003ca class=\"FNLink\" href=\"#Fn2\" id=\"#FNLinkFn2\"\u003e\u003c/a\u003e and post-hoc agnostic models\u003ca class=\"FNLink\" href=\"#Fn3\" id=\"#FNLinkFn3\"\u003e\u003c/a\u003e, into our solution to elucidate the rationale behind AI-driven decision-making. This approach can identify, prioritise, and weigh both maternal protective and risk factors, and quantify their relative impacts on perinatal mental health and well-being outcomes at both group and individual levels. It is anticipated that through the ongoing Aboriginal co-design and cross-disciplinary process, this AI-driven clinical prediction model will support healthcare professionals and Aboriginal women in developing an understanding of and deepened trust in the model's predictions. It will also enhance health professionals\u0026rsquo; confidence in holistic, culturally contextualised clinical decision-making and minimise the cultural biases held by clinicians, and enhance trust and engagement with Aboriginal mothers toward their health professionals\u0026rsquo; supportive care and in their perinatal health service [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe remainder of this paper is organized as follows. Section 2 briefly introduces existing research related to studies using AI techniques for the Aboriginal population and XAI techniques on mental health issues. Section 3 describes the dataset adopted, the machine learning models, and the XAI techniques used for the predictive modelling. The results, along with discussions, are presented in Section 4. Section 5 provides the final discussion and conclusion and suggests future work.\u003c/p\u003e"},{"header":"2. Related Work","content":"\u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eThis section contains two parts. Section \u003cspan refid=\"Sec3\" class=\"InternalRef\"\u003e2.1\u003c/span\u003e provides an overview of AI development to support the health care of the Aboriginal population. Section \u003cspan refid=\"Sec4\" class=\"InternalRef\"\u003e2.2\u003c/span\u003e reviews recent research that utilizes XAI techniques in the mental health field, with a specific focus on structured tabular data, which is the same data type encountered in this work.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. AI Applications in Aboriginal Population Healthcare\u003c/h2\u003e \u003cp\u003eWhile there is increasing interest in utilising AI techniques to address diverse healthcare challenges, there remains a notable scarcity of studies within Aboriginal communities, as evidenced by the limited number of studies conducted in recent years. One study aimed to classify ear diseases through the analysis of otoscopic images collected between 2010 and 2018 from Aboriginal children in the Northern Territory, Australia [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Otolaryngologists analysed and classified the images to label the diseases identified in each screening. Subsequently, this labelled dataset was used to train an image classification algorithm employing deep and transfer learning techniques, achieving an accuracy ranging from 74.4\u0026ndash;92.8% [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. J Scheetz, D Koca, M McGuinness, E Holloway, Z Tan, Z Zhu, R O\u0026rsquo;day, S Sandhu, RJ MacIsaac and C Gilfillan [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] developed a convolutional neural network to detect the presence of diabetic retinopathy in participants, including patients from three Aboriginal Medical Services clinics. Overall, the results showed an accuracy of 96.6% in predicting diabetic retinopathy and 87.7% in identifying retinas without the disease. S Chando, JC Craig, L Burgess, S Sherriff, A Purcell, H Gunasekera, S Banks, N Smith, E Banks and S Woolfenden [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] implemented a multimodal logistic regression model to assess risks in the development and well-being of Aboriginal children. The data were obtained through surveys conducted by Health Services in New South Wales, which included demographic, physical, family, and social factors influencing child health development and well-being for model construction. The study revealed that 32% of the subjects were at high developmental risk, 28% at moderate risk, and 40% at low or no developmental risk.\u003c/p\u003e \u003cp\u003eOn the other hand, little has been accomplished using AI techniques to support health improvements with Aboriginal populations. A recent systematic review of 13 studies on cardiometabolic diseases in primary healthcare highlighted the notable absence of studies employing AI to address heart-related and metabolic diseases within Aboriginal communities [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. However, existing machine learning models, typically trained on data from the general population, may not be suitable for the Aboriginal populations. Historically, Aboriginal peoples have been subjected to mainstream health related \u0026lsquo;solutions\u0026rsquo; with little to no consideration given to their cultural and social determinants of health, or poor access to primary health care and potential biases in the data. Furthermore, many existing AI-assisted decision-support approaches prioritise enhancing prediction accuracy, often leading to increased model complexity due to intricate mathematical modelling [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. However, these approaches often lack transparency in rationales and explanations for the decision-making. Applying existing AI models in Aboriginal communities may have grave risks of data biases, objectification and further exacerbation of health inequality. We propose co-designing an AI model by integrating Aboriginal lived experiences, knowledge and wisdom with professionals\u0026rsquo; experiences and XAI techniques. This approach will ensure the development and deployment of AI technologies align with Aboriginal community desires, needs expectations, while empowering clinicians to make culturally informed and safe, evidence-based decisions via trustworthy AI tools.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. XAI for Mental Health with Tabular Data\u003c/h2\u003e \u003cp\u003eMany AI models are often regarded as \u0026ldquo;black boxes\u0026rdquo; due to their lack of interpretability and transparency [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. To address this issue, XAI has emerged as a promising area, offering easy-to-understand explanations of the predictions and decision-making processes of AI and ML algorithms [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. In mental health applications, understanding the inner workings of AI models becomes crucial. Decision-making in mental health can be highly complex and influenced by various circumstances, making transparency essential for stakeholders [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. N Jothi and W Husain [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] implemented the Shapley value under the proposed method as the feature selection of the data mining classifier on mental health data to predict generalised anxiety disorder among women. This enhanced prediction solution aimed to provide help to any women perceived to be suffering from the disorder. M Ikram, NF Shaikh, JK Vishwanatha and U Sambamoorthi [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] applied eXtreme Gradient Boosting and SHapley Additive exPlanations (SHAP) on survey data, including demographic and socio-economic information, to identify predictors of poor mental health and explain their associations with poor mental health. AU Hussna, II Trisha, IJ Ritun and MGR Alam [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] investigated the impact on students\u0026rsquo; mental health during the COVID-19 pandemic. They identified the top influential variables impacting model output based on SHAP values, including gender, sleep disturbance, perceived discrimination, sitting hours, and stress perception. Similarly, V Adarsh and GR Gangadharan [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] employed SHAP and Local Interpretable Model-agnostic Explanations (LIME) to evaluate IT workers at risk of mental illness, detect early signs of depressive symptoms and use them for training the prediction model. By leveraging common explainable techniques like SHAP and LIME, we can not only enhance predictive precision but also provide clear justifications on the decision-making process, which are imperative for informed clinical decision-making and the formulation of therapeutic interventions [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. This aspect also becomes crucial for fostering trust and ensuring the deployment of AI systems that are suitably sensitive and effective within the mental health domain and targeted groups.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Methods","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Data Source\u003c/h2\u003e \u003cp\u003eThe dataset is drawn from deidentified data generated by the digital assessments that were completed during the WA BCYR pilot [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. This culturally safe and strengths-based perinatal mental-health and social and emotional well-being assessment generates an automated clinical event summary (PDF) which serves as an individualised follow-up management plan for the mother and health professional.\u003c/p\u003e \u003cp\u003eThe de-identified dataset emerged from 293 Aboriginal pilot participants (mothers) between September 9th, 2021, and June 16th, 2023, from one of eight perinatal health-care sites in six services across metropolitan Perth and regional WA. Each participant\u0026rsquo;s data includes 345 variables recorded across different inquiry domains, encompassing strengths and culturally protective factors, common life events, worries, quality of relationships, childhood experiences, family and domestic violence (FDV), tobacco and AOD use. The indicator for maternal distress is derived from the Kessler-5 item psychological distress scale (K5) [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Data pre-processing\u003c/h2\u003e \u003cp\u003eFive records were excluded from the dataset due to missing information regarding the perinatal mental health condition indicator. On the K5 assessment instrument, each item was measured using a five-point scale ranging from 1 to 5. All K5 items are negatively keyed. A score of between 5 to 25 was possible, less than 12 was considered low risk (0); a score of 12 or higher was considered high risk (1) [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. A comprehensive pre-processing and data cleaning procedure was followed by feature selection process that utilised the variable importance measure computed by the random forest algorithm. The algorithm was configured with 500 trees, and the \u0026ldquo;mtry\u0026rdquo; parameter was set to the square root of the total number of variables, rounded down to the nearest integer. Initially, the top 31 most significant variables were selected. Subsequently, these variables were reviewed by the BCYR research team, which included Aboriginal researchers as well as Aboriginal and non-Aboriginal healthcare professionals and users (mothers) of the BCYR digital assessment. By combining their advice based on their domain knowledge and experience, the final list was narrowed down to 20 variables for analysis, in addition to one output variable (K5) to be predicted. The final selected variable information is listed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSelected variables for the prediction model construction.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCode\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQuestion\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeature label\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAnswer options\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q225\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eI feel lonely like I don\u0026rsquo;t belong or fit in\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeeling Lonely\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5: Almost always\u003c/p\u003e \u003cp\u003e4: Often\u003c/p\u003e \u003cp\u003e3: Sometimes\u003c/p\u003e \u003cp\u003e2: A little\u003c/p\u003e \u003cp\u003e1: Hardly ever\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q227\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eI blame myself when things go wrong\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBlaming Herself\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5: Almost always\u003c/p\u003e \u003cp\u003e4: Often\u003c/p\u003e \u003cp\u003e3: Sometimes\u003c/p\u003e \u003cp\u003e2: A little\u003c/p\u003e \u003cp\u003e1: Hardly ever\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q231\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRecently I feel like life is not worth living\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLife Not Worth Living\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1: Never\u003c/p\u003e \u003cp\u003e2: Rarely\u003c/p\u003e \u003cp\u003e3: Sometimes\u003c/p\u003e \u003cp\u003e4: Often\u003c/p\u003e \u003cp\u003e5: Almost always\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q214\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eI feel strong about being a mum\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStrong Mum\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1: Almost always\u003c/p\u003e \u003cp\u003e2: Often\u003c/p\u003e \u003cp\u003e3: Sometimes\u003c/p\u003e \u003cp\u003e4: A little\u003c/p\u003e \u003cp\u003e5: Hardly ever\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q562\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHow likely is it that you will do your goals?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGoal Likely\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1: A lot\u003c/p\u003e \u003cp\u003e2: A fair amount\u003c/p\u003e \u003cp\u003e3: A little bit\u003c/p\u003e \u003cp\u003e4: Not at all\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q904\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eManaging day to day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eManaging Day-to-Day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: Manage well\u003c/p\u003e \u003cp\u003e1: Struggle a bit\u003c/p\u003e \u003cp\u003e2: Struggle a lot\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q534\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClient agrees to making a plan to keep safe to deal with the safety worries\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eKeeping Safety Plan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: Does not agree\u003c/p\u003e \u003cp\u003e1: Client agrees\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q455\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAre there ever times when gambling bothers you?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBothered by Gambling\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: Never\u003c/p\u003e \u003cp\u003e1: Sometimes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q228\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eI make my family proud\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMakes Family Proud\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1: Almost always\u003c/p\u003e \u003cp\u003e2: Often\u003c/p\u003e \u003cp\u003e3: Sometimes\u003c/p\u003e \u003cp\u003e4: A little\u003c/p\u003e \u003cp\u003e5: Hardly ever\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q454\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAre there people close to you gambling?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFamily Gambles\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: No\u003c/p\u003e \u003cp\u003e1: Sometimes\u003c/p\u003e \u003cp\u003e2: Yes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q664\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHow many of these children are in your care?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eChildren in Her Care\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: 0\u003c/p\u003e \u003cp\u003e1: 1\u003c/p\u003e \u003cp\u003e2: 2\u003c/p\u003e \u003cp\u003e3: 3\u003c/p\u003e \u003cp\u003e4: 4 or more\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q450\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHave you smoked cigarettes?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSmoking Cigarettes in Pregnancy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: no\u003c/p\u003e \u003cp\u003e1: sometimes\u003c/p\u003e \u003cp\u003e2: yes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q661\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIs this your first pregnancy?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFirst Pregnancy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: No\u003c/p\u003e \u003cp\u003e1: Yes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q909\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDo you have troubles sleeping?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTrouble Sleeping\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1: Sleeping well\u003c/p\u003e \u003cp\u003e2: Trouble sleeping (not due to pregnancy/baby)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q922\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSecure housing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeed Help with Housing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: No\u003c/p\u003e \u003cp\u003e1: Yes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q663\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHow many previous births have you had?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrevious Births\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: 0\u003c/p\u003e \u003cp\u003e1: 1\u003c/p\u003e \u003cp\u003e2: 2\u003c/p\u003e \u003cp\u003e3: 3\u003c/p\u003e \u003cp\u003e4: 4 or more\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAre you feeling worried?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeeling Worried\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: No\u003c/p\u003e \u003cp\u003e1: Yes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q653\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTold partner/husband about pregnancy?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTold Partner/Husband\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: No\u003c/p\u003e \u003cp\u003e1: Yes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q204\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIs your male partner angry or controlling?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePartner Angry/Controlling\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: No\u003c/p\u003e \u003cp\u003e1: Yes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003efs1.Q195\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIs your male partner moody?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePartner Moody\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0: No\u003c/p\u003e \u003cp\u003e1: Yes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eK5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePsychological distress score category\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003elow_risk\u003c/p\u003e \u003cp\u003ehigh_risk\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Machine Learning Prediction Model\u003c/h2\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1. Traditional ML models\u003c/h2\u003e \u003cp\u003eThree primary machine learning algorithms\u0026mdash;Random Forest (RF), K-Nearest Neighbors (KNN) [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e], and Support Vector Machines (SVM) [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e] \u0026mdash;were selected for training and evaluating the prediction model. These choices were made based on their widespread use and proven effectiveness in mental health research, particularly when dealing with tabular data from relatively small datasets [\u003cspan additionalcitationids=\"CR41\" citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. RF is a classic ensemble learning technique that builds multiple decision trees during training and combines their decisions to make predictions. This process, while effective, is often seen as opaque because it integrates knowledge learned from multiple individual weak classifiers, making it a black-box model. SVM, another widely used black-box algorithm [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], aims to find the optimal hyperplane that separates different classes with a maximal margin. Its effectiveness lies in its ability to classify complex, high-dimensional, and non-linear relationships within the data. KNN, in contrast, is an instance-based learning algorithm that classifies a data point based on the classifications of its neighbours [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. While KNN may offer some interpretability, considering that it can explain predictions based on the proximity of similar instances using features, his reliance on proximity can be heavily influenced by parameter choices, such as the value of K. Moreover, in high-dimensional spaces, the curse of dimensionality may cause nearest neighbours to lose representativeness, diminishing the effectiveness of proximity-based explanations. Also, KNN's explainability could also fall short of providing a clear understanding of the underlying relationships within the data, which is often crucial information for clinical research. Therefore, there remains a need for further explanation techniques to enhance its interpretability.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2. Glassbox ML model\u003c/h2\u003e \u003cp\u003eIn addition to the machine learning models mentioned above, we also employed Explainable Boosting Machines (EBM) [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e] for the prediction model. Unlike the previously discussed models, EBM is a glassbox model, which inherently provides explanations for prediction outcomes concerning predictor features, enabling us to interpret and comprehend the model's decisions. Specifically, EBM is a tree-based, gradient boosting Generalized Additive Model (GAM) with automatic pairwise interaction identification [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. The contribution of each feature to the final prediction can be visualised and sorted, providing an explanation. The final EBM model is a sum of one-dimensional models for each feature and two-dimensional models for potential pairwise interaction terms. This sum is then passed through a logistic function in the case of classification tasks to ensure the output lies between 0 and 1, representing probabilities of the positive class. In general, EBM can be represented by the following form [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]:\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:g\\left(E\\right[y\\left]\\right)={\\beta\\:}_{0}+\\sum\\:{f}_{i}\\left({x}_{i}\\right)+\\sum\\:{f}_{ij}({x}_{i},{x}_{j})\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:g(\\cdot\\:)\\)\u003c/span\u003e\u003c/span\u003e is link function, which transforms the expected value of the target variable \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y\\)\u003c/span\u003e\u003c/span\u003e depending on the classification or regression problem. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\beta\\:}_{0}\\)\u003c/span\u003e\u003c/span\u003e is the intercept, representing the expected value of the prediction outcome when all the features are equal to zero. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{f}_{i}\\left({x}_{i}\\right)\\:\\)\u003c/span\u003e\u003c/span\u003eis a learned function based on tree-based gradient boosting for each individual feature \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{i}\\)\u003c/span\u003e\u003c/span\u003e, indicating its contribution to the prediction outcome. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{f}_{ij}({x}_{i},{x}_{j})\\)\u003c/span\u003e\u003c/span\u003e is another learned function based on tree-based methods for a pair of features \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:({x}_{i},{x}_{j})\\)\u003c/span\u003e\u003c/span\u003e, modelling their features\u0026rsquo; interactions on the prediction [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. To make individual prediction, each function \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{f}_{i}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{f}_{ij}\\)\u003c/span\u003e\u003c/span\u003e act as a lookup table per feature and two-dimensional features, respectively, returning a term contribution. These term contributions are then simply added up and passed through the link function \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:g(\\cdot\\:)\\)\u003c/span\u003e\u003c/span\u003e for the final prediction.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e3.3.3. Prediction performance evaluation\u003c/h2\u003e \u003cp\u003eTo ensure a fair and comparable learning and evaluation process, we employed 10-fold cross-validation. In simpler terms, 10-fold cross-validation involves dividing the designated dataset into ten subsets. The model is then trained on nine of these subsets and validated on the remaining one. This process repeats, cycling through all ten subsets for testing [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. In this study, 10-fold cross-validation was utilised for hyperparameter tuning with grid search and performance evaluation. The reported performance of the trained models aggregates the results from 10-fold cross-validation, using the optimal model with the hyperparameter values selected through the model selection procedure. Each model is evaluated using an extensive set of metrics, including accuracy, precision, recall, F1-Score, and Area Under the Curve (AUC), which are common metrics for evaluating classification performance, particularly in health risk assessment prediction tasks. Additionally, we presented the 95% confidence intervals for each performance metric, measuring the uncertainty surrounding them.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Post-hoc Agnostic Explanation Techniques\u003c/h2\u003e \u003cdiv id=\"Sec13\" class=\"Section3\"\u003e \u003ch2\u003e3.4.1. SHAP (Shapley Additive exPlanations)\u003c/h2\u003e \u003cp\u003eSHAP is a model-agnostic technique designed to elucidate the outputs of black-box ML models at both individual and global levels. Rooted in cooperative game theory, SHAP computes Shapley values to determine a fair distribution of the prediction among the features of a given data instance. The Shapley value explanation is represented as an additive feature attribution method using a linear model. The SHAP model can be represented as follows [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]:\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:g\\left({z}^{{\\prime\\:}}\\right)={\\varphi\\:}_{0}+\\sum\\:_{j=1}^{M}{\\varphi\\:}_{j}{z}_{j}^{{\\prime\\:}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:g\\)\u003c/span\u003e\u003c/span\u003e is the model, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{z}^{{\\prime\\:}}\\in\\:{\\left\\{0,\\:1\\right\\}}^{M}\\)\u003c/span\u003e\u003c/span\u003e is the coalition vector, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:M\\)\u003c/span\u003e\u003c/span\u003e is the maximum coalition size, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\varphi\\:}_{j}\\in\\:\\mathcal{\\:}\\mathcal{R}\\)\u003c/span\u003e\u003c/span\u003e is the feature attribution for feature \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:j\\)\u003c/span\u003e\u003c/span\u003e, representing Shapley values. The larger the absolute Shapley values, the more important the associated features. Furthermore, Shapley values can be aggregated into global explanations. After running SHAP for every instance, a matrix of Shapley values, with one vector row per data instance and one column vector per feature, can be formulated. Analysing this Shapley value matrix enables us to interpret the entire model at the global level. Kernel SHAP is a specific technique for approximating Shapley values through a sampling process, with the sampled instances weighted using the SHAP kernel based on the coalition\u0026rsquo;s weight in the Shapley value estimation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e3.4.2. LIME (Local Interpretable Model-Agnostic Explanations)\u003c/h2\u003e \u003cp\u003eLIME is a widely adopted model-agnostic technique designed to elucidate individual predictions by constructing surrogate local models that approximate the output of a black-box model, thus offering the capacity for local interpretability [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]. This technique tests how the machine learning model's predictions change in response to variations in the input data. Specifically, LIME generates a perturbed dataset based on the individual instance of interest and its corresponding predictions from the black-box model. Subsequently, LIME utilizes this perturbed dataset with proximity weightings to train an interpretable model to approximate the black-box model's decision-making process. The LIME local surrogate model optimization problem can be represented as follows.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabb\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\underset{g\\in\\:G}{\\text{min}}L\\:\\left(f,\\:g,\\:{\\pi\\:}_{\\varvec{x}}\\right)+{\\Omega\\:}\\left(g\\right)\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe explanation model \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:g\\)\u003c/span\u003e\u003c/span\u003e for the individual instance \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{x}\\)\u003c/span\u003e\u003c/span\u003e of interest can be a simple interpretable model, e.g., linear regression or a decision tree, that minimize the square loss function \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:L\\)\u003c/span\u003e\u003c/span\u003e, which measures how close the prediction of the explanation model is to the prediction of the original black-box model \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:f\\)\u003c/span\u003e\u003c/span\u003e. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\pi\\:}_{\\varvec{x}}\\)\u003c/span\u003e\u003c/span\u003e is the proximity measure deciding the perturbed instance neighbourhood around instance \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{x}\\)\u003c/span\u003e\u003c/span\u003e that are used for the explanation. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\Omega\\:}\\left(g\\right)\\)\u003c/span\u003e\u003c/span\u003e is the regularisation term to control the model complexity.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e3.4.3. PDP (Partial Dependence Plots)\u003c/h2\u003e \u003cp\u003ePDP is a global explanatory, model-agnostic visualisation technique that displays the marginal effect of one or two features on the predicted outcome of a black-box ML model while holding all other features constant [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. This technique allows for an intuitive understanding of the relationship between the input features and the model's predictions, especially for non-technical background stakeholders seeking insights into the model's behaviour. Specifically, first, users choose a feature of interest. Once a range of values for the chosen feature is established, based on the user's interest in observing partial dependence, the model computes predictions for each value within that specified range. Simultaneously, all other features are held constant, either fixing them at specific values or by using their average values. Finally, the predicted outcomes against the values of the selected feature are plotted, resulting in a graphical representation of how changes in that specific feature influence the model\u0026rsquo;s predictions.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Results and Discussion","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Prediction Model Performance\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e displays the training and testing results for various performance metrics on the BCYR dataset. The best results are highlighted in bold. It is evident that among all the black-box ML models, RF outperformed with an accuracy of 0.829, an F1 score of 0.736, and an AUC of 0.795. Moreover, the glassbox model EBM exhibited the highest performance among all the models, achieving an accuracy of 0.849, an F1 score of 0.771, and an AUC of 0.821, indicating improved predictive capabilities compared to other black-box models. KNN and SVM demonstrated comparable accuracy, slightly below 0.8 (0.798, 0.794), and comparable AUC (0.733, 0.742). In terms of precision and recall metrics, KNN exhibited the highest precision at 0.868 but the lowest recall at 0.514, resulting in an unsatisfactory F1 score of 0.621. Such significant discrepancies in high precision and low recall may potentially miss a considerable number of high-risk cases in the context of mental health risk detection. On the other hand, EBM and RF maintain a commendable balance between precision and recall. Specifically, EBM achieved a precision of 0.829, a recall of 0.727, and an F1 score of 0.771, while RF attained a precision of 0.820, a recall of 0.680, and an F1 score of 0.736. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e plots the receiver operating characteristic curve (ROC) curves of all the models on one set of testing. It visually shows that the EBM and RF models are the best AUC performers with values of 0.814 and 0.781, respectively, indicating better performance in distinguishing between high-risk and low-risk cases. In summary, EBM and RF outperformed among all the classifiers in achieving high accuracy, F1 score, and AUC, ensuring a reasonable level of both precision and recall for the prediction task, with EBM achieving the best performance.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformances of all machine learning models for predicting perinatal mental health risk.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"12\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAccuracy Mean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSTD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePrecision Mean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSTD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eRecall Mean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSTD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eF1 Mean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eSTD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eAUC Mean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e \u003cp\u003eSTD\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.900\u003c/p\u003e \u003cp\u003e(0.8700, 0.9300)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.914\u003c/p\u003e \u003cp\u003e(0.8826, 0.9462)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.788\u003c/p\u003e \u003cp\u003e(0.7240, 0.8513)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.845\u003c/p\u003e \u003cp\u003e(0.7961, 0.8933)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.874\u003c/p\u003e \u003cp\u003e(0.8370, 0.9119)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTesting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.829\u003c/p\u003e \u003cp\u003e(0.7960, 0.8617)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.8200\u003c/p\u003e \u003cp\u003e(0.7524, 0.8869)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.680\u003c/p\u003e \u003cp\u003e(0.6099, 0.7501)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.736\u003c/p\u003e \u003cp\u003e(0.6859, 0.7851)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.795\u003c/p\u003e \u003cp\u003e(0.7581, 0.8318)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003cp\u003e(1.000, 1.000)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003cp\u003e(1.000, 1.000)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003cp\u003e(1.000, 1.000)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003cp\u003e(1.000, 1.000)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003cp\u003e(1.000, 1.000)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTesting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.798\u003c/p\u003e \u003cp\u003e(0.7524, 0.8430)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.868\u003c/p\u003e \u003cp\u003e(0.7860, 0.9491)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.514\u003c/p\u003e \u003cp\u003e(0.3846, 0.6426)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.621\u003c/p\u003e \u003cp\u003e(0.5204, 0.7216)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.733\u003c/p\u003e \u003cp\u003e(0.6709, 0.7948)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.886\u003c/p\u003e \u003cp\u003e(0.8797, 0.8927)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.899\u003c/p\u003e \u003cp\u003e(0.8882, 0.9091)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.764\u003c/p\u003e \u003cp\u003e(0.7505, 0.7770)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.826\u003c/p\u003e \u003cp\u003e(0.8152, 0.8359)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.858\u003c/p\u003e \u003cp\u003e(0.8506, 0.8661)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTesting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.849\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(0.8170, 0.8814)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.829\u003c/p\u003e \u003cp\u003e(0.7689, 0.8900)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.727\u003c/p\u003e \u003cp\u003e(0.6599, 0.7946)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e\u003cb\u003e0.771\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(0.7169, 0.8245)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e\u003cb\u003e0.821\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(0.7829, 0.8593)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.858\u003c/p\u003e \u003cp\u003e(0.8275, 0.8894)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.920\u003c/p\u003e \u003cp\u003e(0.8943, 0.9466)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.652\u003c/p\u003e \u003cp\u003e(0.5804, 0.7236)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.760\u003c/p\u003e \u003cp\u003e(0.7055, 0.8150)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.812\u003c/p\u003e \u003cp\u003e(0.7715, 0.8517)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTesting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.794\u003c/p\u003e \u003cp\u003e(0.7514, 0.8373)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.805\u003c/p\u003e \u003cp\u003e(0.7195, 0.8906)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.570\u003c/p\u003e \u003cp\u003e(0.4604, 0.6796)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.650\u003c/p\u003e \u003cp\u003e(0.5675, 0.7335)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.742\u003c/p\u003e \u003cp\u003e(0.6880, 0.7966)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Explanation Results\u003c/h2\u003e \u003cdiv id=\"Sec19\" class=\"Section3\"\u003e \u003ch2\u003e4.2.1. Global interpretation by EBM and SHAP\u003c/h2\u003e \u003cp\u003eAs EBM demonstrates high predictive performance and operates as a transparent glassbox model, we generated the explanation from EBM and illustrated the global feature importance over the whole data set in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Longer bars in the figure indicate higher importance of features in the model's predictions. It is noteworthy that specific features exert a more significant influence on the decision-making process of the EBM model. For instance, the feature \u0026ldquo;Feeling Lonely\u0026rdquo; emerges as the most impactful, suggesting that a participant\u0026rsquo;s response to a reflection concerning feelings of loneliness, might serve as a robust predictor for perinatal mental health risk. This is followed by \u0026ldquo;Blaming Herself\u0026rdquo; and \u0026ldquo;Makes Family Proud,\u0026rdquo; indicating that these two features have a higher contribution to the overall prediction (risk or protective). Other features, such as \u0026ldquo;Managing Day-to-Day\u0026rdquo; and \u0026ldquo;Strong Mum,\u0026rdquo; also demonstrate a notable impact in relation to the K5 target outcome, which is aligned with previous study findings [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Moreover, the figure shows specific multi-dimensional interactions that add increased weighting to the model\u0026rsquo;s predictions. For example, the feature \u0026ldquo;Feeling Lonely\u0026rdquo; in combination with features such as \u0026ldquo;Life Not Worth Living,\u0026rdquo; \u0026ldquo;Blaming Herself,\u0026rdquo; or \u0026ldquo;Partner Angry/Controlling\u0026rdquo; places greater predictive power. Likewise, the joint effects of \u0026ldquo;Blaming Herself\u0026rdquo; and \u0026ldquo;Family Gambles\u0026rdquo; or \u0026ldquo;Makes Family Proud\u0026rdquo; are also identified as important feature combinations affecting the model predictions.\u003c/p\u003e \u003cp\u003eFor the high-performing black-box RF model, we utilised SHAP to also gain insights into the model\u0026rsquo;s explanation. We displayed the global feature importance ranking in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The most important features, including \u0026ldquo;Feeling Lonely,\u0026rdquo; \u0026ldquo;Blaming Herself,\u0026rdquo; \u0026ldquo;Managing Day-to-Day,\u0026rdquo; \u0026ldquo;Life Not Worth Living,\u0026rdquo; and \u0026ldquo;Makes Family Proud,\u0026rdquo; exactly overlap with the top features selected from EBM. Additionally, features related to questions concerning \u0026ldquo;Family Gambles,\u0026rdquo; \u0026ldquo;Strong Mum,\u0026rdquo; \u0026ldquo;Goal Likely,\u0026rdquo; and \u0026ldquo;Trouble Sleeping\u0026rdquo; were subsequently ranked, aligning with EBM\u0026rsquo;s ranking tier as well.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section3\"\u003e \u003ch2\u003e4.2.2. Local interpretation by EBM, SHAP, LIME\u003c/h2\u003e \u003cp\u003eEBM can also provide an interpretation for the model\u0026rsquo;s prediction on an individual instance. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows one Aboriginal woman who is low risk, labelled \u0026ldquo;Instance I,\u0026rdquo; it shows the specific contributions of each feature and their interactions in combination toward risk prediction for this individual by EBM. EBM accurately predicts this individual as \u0026ldquo;low risk\u0026rdquo; (class\u0026thinsp;=\u0026thinsp;0), with a high probability score of 0.904. The y-axes of the plot show impactful features and their corresponding values in brackets. Specifically, contributing features highlighted in blue, such as \u0026ldquo;Blaming Herself\u0026rdquo; = 1.00 (hardly ever) and \u0026ldquo;Feeling Lonely\u0026rdquo; = 1.00 (hardly ever), contribute to shifting the model's prediction towards the low-risk category. These two features stand out, suggesting that hardly ever feeling lonely and hardly ever blaming herself are strongly protective factors for this Aboriginal mother. Additionally, the combination of \u0026ldquo;Blaming Herself\u0026rdquo; = 1.00 (hardly ever) and \u0026ldquo;Family Gambles\u0026rdquo; = 0.00 (loved ones do not gamble) shows an additive positive influence, pushing the prediction towards low risk. Conversely, contributing features displayed in orange in the figure, would have pushed the model's prediction away from low risk. For example, \u0026ldquo;Children in Her Care\u0026rdquo; = 3 (having three children in care) and \u0026ldquo;Strong Mum\u0026rdquo; = 3 (sometimes), and \u0026ldquo;Makes Family Proud\u0026rdquo; = 2 (often) and \u0026ldquo;Partner Moody\u0026rdquo; = 1 (yes), are risk factors that increase this woman\u0026rsquo;s risk. We noted that \u0026ldquo;Makes Family Proud\u0026rdquo; and \u0026ldquo;Strong Mum\u0026rdquo; were highly protective only when women selected the top rating: \u0026lsquo;Always\u0026rsquo;.\u003c/p\u003e \u003cp\u003eLooking at the SHAP techniques, we generated force plots in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e to visualise the contributions of important features to the same (\u0026ldquo;Instance I\u0026rdquo;) woman\u0026rsquo;s prediction. The lower prediction score of 0.12 indicates that this participant is at low risk. The key features influencing the prediction for this individual are depicted in red and blue. Red indicates features that elevated the model's score towards high risk. Blue signifies features that reduced the risk. Features having greater impact on prediction scores are located closer to the dividing boundary between red and blue in the figure. Therefore, the strongly protective factors contributing to this low-risk prediction, including \u0026ldquo;hardly ever blames herself,\u0026rdquo; \u0026ldquo;hardly ever feels lonely,\u0026rdquo; and \u0026ldquo;close loved ones do not gamble,\u0026rdquo; align with the EBM\u0026rsquo;s individual interpretation. Additionally, \u0026ldquo;sometimes feels strong about being a mum\u0026rdquo; and \u0026ldquo;moody partner\u0026rdquo; tended to push the prediction towards higher risk, which again is similar to EBM\u0026rsquo;s interpretation.\u003c/p\u003e \u003cp\u003eWe employed an additional technology, the LIME technique, to further explore individual influencing features for the same woman. Overall, the results were very similar to EBM and SHAP technologies. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e illustrates that the most influential features in predicting low risk were \u0026ldquo;hardly ever feels lonely,\u0026rdquo; \u0026ldquo;hardly ever blames herself,\u0026rdquo; \u0026ldquo;never feels life is not worth living,\u0026rdquo; and \u0026ldquo;close loved ones do not gamble.\u0026rdquo; Interestingly LIME uniquely identified that \u0026ldquo;needs no help with housing\u0026rdquo; was a contributor to low risk, not identified by EBM and SHAP.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e provides the EBM\u0026rsquo;s local explanation for another Aboriginal woman labelled \u0026ldquo;Instance II,\u0026rdquo; who was at greater risk. It was accurately predicted with a high risk of perinatal mental health problems and a probability score of 0.655. The chart reveals that the factors significantly influencing the prediction decision towards high risk include \u0026ldquo;Feeling Lonely\u0026rdquo;=2 (feels a little lonely), \u0026ldquo;Trouble Sleeping\u0026rdquo;=2 (having trouble sleeping), \u0026ldquo;Family Gambles\u0026rdquo;=2 (close loved ones gamble), \u0026ldquo;Makes Family Proud\u0026rdquo;=3 (sometimes makes her family proud), and \u0026ldquo;Goal Likely\u0026rdquo;=2 (a fair amount). Meanwhile, features in blue, such as \u0026ldquo;Blaming Herself\u0026rdquo;=1 (hardly ever blames herself), \u0026ldquo;Managing Day-to-Day\u0026rdquo;=0 (manages day-to-day well), \u0026ldquo;Life Not Worth Living\u0026rdquo;=1 (never feels life is not worth living), \u0026ldquo;Strong Mum\u0026rdquo;=1 (almost always feels strong about being a mum), \u0026ldquo;Children in Her Care\u0026rdquo;=1 (having one child in care), and \u0026ldquo;Previous Births\u0026rdquo;=1 (one previous birth), serve as protective factors in this specific instance.\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e shows the prediction interpretation for the same individual using SHAP. A consistent group of answers to the questions that drive the prediction towards a high risk was identified, including \u0026ldquo;Family Gambles,\u0026rdquo; \u0026ldquo;Feeling Lonely,\u0026rdquo; \u0026ldquo;Makes Family Proud,\u0026rdquo; \u0026ldquo;Goal Likely,\u0026rdquo; and \u0026ldquo;Trouble Sleeping,\u0026rdquo; although the significance level ranking shows slight differences. Meanwhile, not \u0026ldquo;Blaming Herself\u0026rdquo; and \u0026ldquo;Managing Day-to-Day\u0026rdquo; were identified as mitigating factors that attempt to reduce this high-risk prediction.\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e provides LIME\u0026rsquo;s local interpretation of the same individual. It selected the same group of risk factors as SHAP, except for \u0026ldquo;Trouble Sleeping,\u0026rdquo; which was not chosen by LIME but was selected by both SHAP and EBM. The protective factors also largely align with the other two methods, except for \u0026ldquo;Need Help With Housing\u0026rdquo; =0, chosen as a protective factor by LIME, which was not picked as a top protective factor by the other two methods.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003e4.2.3. High influential factor interpretation by PDP\u003c/h2\u003e \u003cp\u003eThrough interpretation conducted using both glassbox and model-agnostic techniques, we identified two highly significant questions impacting the prediction outcomes: 'Blaming Herself' and 'Feeling Lonely.' We further adopted their corresponding PDP plots to reveal their individual relationships with the average predicted outcome shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e. In the case of 'Blaming Herself,' mothers who hardly ever blame themselves are associated with a low risk. However, starting from 'A Little' and progressing to 'Sometimes,' 'Often,' and 'Almost Always,' there is a significant increase in the predicted risk of perinatal mental health issues. This suggests that as the frequency of self-blame rises, even starting from 'A Little,' the associated risk shows a significant increasing trend. There is not much difference in the impact between the categories 'Sometimes,' 'Often,' and 'Almost Always,' as they all show a significant level of relation with high risk. Similar observations were made for the case of 'Feeling Lonely,' where mothers who reported 'Hardly Ever' feeling lonely are associated with a low risk.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe further generated a PDP plot to visualise the interplay between these two significant factors. The heatmap colour gradient represents the predicted risk level, with light cream indicating a low risk (closer to 0) and purple representing a high risk (closer to 1). Mothers who reported \u0026ldquo;Hardly Ever\u0026rdquo; blaming themselves and \u0026ldquo;Hardly Ever\u0026rdquo; feeling lonely are in the lightest zone, suggesting the lowest predicted risk. However, the deeper colour zone is pronounced for respondents who reported feeling lonely starting from \u0026ldquo;A Little\u0026rdquo; and beyond and blaming themselves starting from \u0026ldquo;A Little\u0026rdquo; and beyond. This combined emotional state of frequent loneliness and self-blaming puts them at a much higher predicted risk for perinatal mental health issues. Regardless of whether the response was \u0026ldquo;Sometimes,\u0026rdquo; \u0026ldquo;Often,\u0026rdquo; or \u0026ldquo;Always\u0026rdquo; for loneliness and self-blame, it led to the highest level of risk, with almost the same effect.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"5. CONCLUSIONs AND FINAL REMARKS","content":"\u003cp\u003eIn conclusion, our results reveal that both the glassbox EBM model and the blackbox RF model, which are ensemble-based methods, excel in predicting perinatal mental health issues among Aboriginal mothers. The EBM achieved an accuracy of 0.849 and an AUC of 0.821, while the random forest model achieved an accuracy of 0.829 and an AUC of 0.795. Additionally, we compared the explainability generated by EBM and post-hoc agnostic XAI technologies including SHAP and LIME at both the global and individual levels. Generally, both EBM and SHAP provide global explanations that highlight questions about \u0026ldquo;Feeling Lonely\u0026rdquo; and \u0026ldquo;Blaming Herself\u0026rdquo; as dominant factors influencing the overall prediction outcome. Additionally, items regarding \u0026ldquo;Makes Family Proud,\u0026rdquo; \u0026ldquo;Life Not Worth Living,\u0026rdquo; \u0026ldquo;Managing Day-to-Day,\u0026rdquo; \u0026ldquo;Strong Mum,\u0026rdquo; and \u0026ldquo;Family Gambles\u0026rdquo; are also identified as top-ranking questions. These show minimal variations across different XAI interpretation methods. At the individual level, we examined two cases\u0026mdash;one high risk and one low risk\u0026mdash;as examples, using EBM, SHAP, and LIME to generate visual maps. These maps clearly displayed significant protective and risk factors contributing to the prediction outputs, with very minor variations in the different XAI interpretation methods. Across all three XAI interpretation methods, emotional states related to feelings of loneliness, self-blame, and perceptions of self-worth consistently emerge as powerful predictors for perinatal mental health issues for Aboriginal mothers. Additionally, certain features in combination such as 'Feeling Lonely' and 'Life Not Worth Living,' also influenced predictions of perinatal maternal risk.\u003c/p\u003e \u003cp\u003eBy analysing the combination of global and local insights gained from these XAI methods, we can achieve a better and more transparent understanding of the data-driven model's decision-making process regarding Aboriginal mothers' perinatal mental health and associated factors, and even delve into individual-level insights. We anticipate that this XAI-assisted and co-designed prediction model can enhance clinicians' and health services' capacity and confidence to provide culturally sensitive clinical reasoning and accurate interpretation of screening results that reduces existing biases. We predict it may effectively facilitate early and accurate identification of perinatal mental-health issues, enhance relevant and quality supports and referrals, strengthen culturally secure care and minimise unnecessary child protection notifications. Through combining transparent XAI outputs with lived experience wisdom this clinical prediction model may provide new insights, knowledge, and mechanisms for addressing perinatal mental-health issues among Aboriginal women, facilitate closer multidisciplinary collaboration, improve communication, and build trust between new techniques, researchers, clinicians, and Aboriginal community. This would result in continuous improvement in service quality and efficiency, cost saves and support Aboriginal families to flourish and thrive.\u003c/p\u003e \u003cp\u003eThere are limitations that need to be addressed in future work. Firstly, the sample size used for building the prediction model is limited, despite the high number of features in the de-identified data. As more data is collected through the BCYR program across different regions, future opportunities will arise to increase the sample size, enhancing the model's performance. Secondly, the current visual maps generated from the XAI techniques will be improved through the co-design process to enhance readability and explainability, especially for Aboriginal women and clinicians. This user-friendly and culturally sensitive visual prediction model will ensure accurate and efficient interpretation by all stakeholders involved.\u003c/p\u003e"},{"header":"Abbreviations","content":" \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabc\" border=\"1\"\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(ACC)\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eAccuracy\u003c/div\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(AD)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eDementia\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(AI)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eArtificial Intelligence\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(AUC)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eArea Under the Curve\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(AUROC)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eArea Under the Receiver Operating Characteristic Curve\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(BCYR)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e\u0026ldquo;Baby Coming You Ready?\u0026rdquo;\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(CMD)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eCardiometabolic Diseases\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(CNN)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eConvolutional Neural Network\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(DNN)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eDeep Neuronal Network\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(EBM)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eExplainable Boosting Machine\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(HER/ EMR)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eElectronic Health Record or Electronic Medical Record\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(IEA)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eImage-emotion Alignment\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(ITA)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eImage-text Alignment\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(LIME)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eLocal Interpretable Model-Agnostic Explanations\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(LR)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eLogistic Regression\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(ML)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eMachine Learning\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(NA)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eNo Answer\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(NB)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eNa\u0026iuml;ve Bayes\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(NLP)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eNatural Language Processing\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(PDP)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003ePartial Dependence Plots\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(PHC)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003ePrimary Health Care\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(PNDA)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003ePerinatal depression\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(PRECIS)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003ePrecision\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(RF)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eRandom Forest\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(ROC)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eReceiver Operating Characteristic\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(SHAP)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eSHapley Additive exPlanations\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003e(XAI)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eExplainable Artificial Learning\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003cbr/\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e \u003cp\u003eEthics approval was obtained from the Human Research Ethics Committee (HREC) - Western Australia Research Governance Service: RGS000000649; Murdoch University: 2021/101; Western Australian Aboriginal Health Ethics Committee (WAAHEC): HREC553. Access to deidentified data for this study made available only from consenting participants. A consent button to either use or not to use deidentified data for research purposes is embedded in the digital application.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003e Not applicable.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting Interests\u003c/h2\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThe work was supported by the Western Australian Future Health Research and Innovation Fund (Grant ID IC2023-GAIA/18), and J. Kotz and G. Wang are supported by the Google Inclusion Research Award.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eG.W., W.K., R.M., R.W., and J.K. designed the work; G.W., W.K., H.B., and J.Q. performed the experiments and data analysis; G.W., J.Q., and J.K. drafted the manuscript; G.W. and J.K. substantively revised the manuscript. All authors interpreted the data and results and reviewed the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eNot Applicable.\u003c/p\u003e\u003ch2\u003eAvailability of data and materials\u003c/h2\u003e\u003cp\u003eThe data supporting the findings of this study are not openly available. Requests for more detailed information can be directed to the corresponding author.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBrummelte S, Galea LA. Postpartum depression: Etiology, treatment and consequences for maternal care. Horm Behav. 2016;77:153\u0026ndash;66.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMitchell AR, Gordon H, Lindquist A, Walker SP, Homer CS, Middleton A, Cluver CA, Tong S, Hastie R. Prevalence of perinatal depression in low-and middle-income countries: A systematic review and meta-analysis. JAMA Psychiatry. 2023;80(5):425.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHummel AD, Ronen K, Bhat A, Wandika B, Choo EM, Osborn L, Batra M, Kinuthia J, Kumar M, Unger JA. Perinatal depression and its impact on infant outcomes and maternal-nurse SMS communication in a cohort of Kenyan women. BMC Pregnancy Childbirth. 2022;22(1):723.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Backer K, Wilson CA, Dolman C, Vowles Z, Easter A. Rising rates of perinatal suicide. 2023, 381:e075414.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAustralian Institute of Health and Welfare. Child Protection Australia 2019-20. Child welfare series no. 74. Cat. no. CWS 78. Australian Institute of Health and Welfare; 2021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIndigenous health and wellbeing. [\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.aihw.gov.au/reports/australias-health/indigenous-health-and-wellbeing]\u003c/span\u003e\u003cspan address=\"https://www.aihw.gov.au/reports/australias-health/indigenous-health-and-wellbeing]\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOwais S, Faltyn M, Johnson AV, Gabel C, Downey B, Kates N, Van Lieshout RJ. The perinatal mental health of indigenous women: A systematic review and meta-analysis. Can J Psychiatry. 2020;65(3):149\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLima F, Shepherd C, Wong J, O\u0026rsquo;Donnell M, Marriott R. Trends in mental health related contacts among mothers of Aboriginal children in Western Australia (1990\u0026ndash;2013): A linked data population-based cohort study of over 40 000 children. BMJ Open. 2019;9(7):e027733.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRatajczak T. Exploring Cultural and Clinical Factors Contributing to Aboriginal and Torres Strait Islander Women\u0026rsquo;s Resilience: A Study Utilising the Baby Coming You Ready Program. WA: Murdoch University; 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKotz J, Bennett E, Reibel T, AhChee V, Marriott R. Baby coming you ready' an innovative screening and assessment tool for perinatal mental health with Aboriginal families: A protocol for pilot evaluation. Australian J Child Family Health Nurs. 2021;18(2):19\u0026ndash;26.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGee G, Dudgeon P, Schultz C, Hart A, Kelly K. Aboriginal and Torres Strait Islander social and emotional wellbeing. In: \u003cem\u003eWorking together: Aboriginal and Torres Strait Islander mental health and wellbeing principles and practice.\u003c/em\u003e Edited by Dudgeon P, Milroy H, Walker R, vol. 2; 2014: 55\u0026ndash;68.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNovick AM, Kwitowski M, Dempsey J, Cooke DL, Dempsey AG. Technology-based approaches for supporting perinatal mental health. Curr Psychiatry Rep. 2022;24(9):419\u0026ndash;29.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBilal AM, Fransson E, Br\u0026auml;nn E, Eriksson A, Zhong M, Gid\u0026eacute;n K, Elofsson U, Axfors C, Skalkidou A, Papadopoulos FC. Predicting perinatal health outcomes using smartphone-based digital phenotyping and machine learning in a prospective Swedish cohort (Mom2B): Study protocol. BMJ Open. 2022;12(4):e059033.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCellini P, Pigoni A, Delvecchio G, Moltrasio C, Brambilla P. Machine learning in the prediction of postpartum depression: A review. J Affect Disord. 2022;309:350\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S, Pathak J, Zhang Y. Using electronic health records and machine learning to predict postpartum depression. MEDINFO 2019: Health and Wellbeing e-Networks for All. IOS; 2019. pp. 888\u0026ndash;92.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Y, Wang S, Hermann A, Joly R, Pathak J. Development and validation of a machine learning algorithm for predicting the risk of postpartum depression among pregnant women. J Affect Disord. 2021;279:1\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhong M, Van Zoest V, Bilal AM, Papadopoulos F, Castellano G. Unimodal vs. multimodal prediction of antenatal depression from smartphone-based survey data in a longitudinal study. In: \u003cem\u003e2022 International Conference on Multimodal Interaction: 2022; New York\u003c/em\u003e; 2022: 455\u0026ndash;467.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJaved F, Gilani SO, Latif S, Waris A, Jamil M, Waqas A. Predicting risk of antenatal depression and anxiety using multi-layer perceptrons and support vector machines. J Personalized Med. 2021;11(3):199.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGarriga R, Mas J, Abraha S, Nolan J, Harrison O, Tadros G, Matic A. Machine learning model to predict mental health crises from electronic health records. Nat Med. 2022;28(6):1240\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKelly B. Culturally Safe Mental Health Screening for Aboriginal Mothers: An exploration of the value of Baby Coming You Ready program in child health settings. WA: Murdoch University; 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHabib A-R, Crossland G, Patel H, Wong E, Kong K, Gunasekera H, Richards B, Caffery L, Perry C, Sacks R. An artificial intelligence computer-vision algorithm to triage otoscopic images from Australian Aboriginal and Torres Strait Islander children. Otology Neurotology. 2022;43(4):481\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScheetz J, Koca D, McGuinness M, Holloway E, Tan Z, Zhu Z, O\u0026rsquo;day R, Sandhu S, MacIsaac RJ, Gilfillan C. Real-world artificial intelligence-based opportunistic screening for diabetic retinopathy in endocrinology and indigenous healthcare settings in Australia. Sci Rep. 2021;11(1):15808.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChando S, Craig JC, Burgess L, Sherriff S, Purcell A, Gunasekera H, Banks S, Smith N, Banks E, Woolfenden S. Developmental risk among Aboriginal children living in urban areas in Australia: The Study of Environment on Aboriginal Resilience and Child Health (SEARCH). BMC Pediatr. 2020;20:1\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJeong K, Mallard AR, Coombe L, Ward J. Artificial intelligence and prediction of cardiometabolic disease: Systematic review of model performance and potential benefits in indigenous populations. Artif Intell Med. 2023;139:102534.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNagendran M, Chen Y, Lovejoy CA, Gordon AC, Komorowski M, Harvey H, Topol EJ, Ioannidis JPA, Collins GS, Maruthappu M. Artificial intelligence versus clinicians: Systematic review of design, reporting standards, and claims of deep learning studies in medical imaging. BMJ (Online) 2020, 368.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLetzgus S, M\u0026uuml;ller K-R. Towards transparent and robust data-driven wind turbine power curve models. arXiv preprint 2023:arXiv:230409835.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHolzinger A, Saranti A, Molnar C, Biecek P, Samek W. Explainable AI methods - A brief overview. In: \u003cem\u003exxAI - Beyond Explainable AI: International Workshop, Held in Conjunction with ICML 2020, July 18, 2020, Vienna, Austria, Revised and Extended Papers.\u003c/em\u003e Edited by Holzinger A, Goebel R, Fong R, Moon T, M\u0026uuml;ller K-R, Samek W. Cham: Springer International Publishing; 2022: 13\u0026ndash;38.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eByeon H. Advances in machine learning and explainable artificial intelligence for depression prediction. Int J Adv Comput Sci Appl 2023, 14(6).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoh HW, Ooi CP, Seoni S, Barua PD, Molinari F, Acharya UR. Application of explainable artificial intelligence for healthcare: A systematic review of the last decade (2011\u0026ndash;2022). \u003cem\u003eComputer Methods and Programs in Biomedicine\u003c/em\u003e 2022, 226:107161.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJothi N, Husain W. Predicting generalized anxiety disorder among women using Shapley value. J Infect Public Health. 2021;14(1):103\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIkram M, Shaikh NF, Vishwanatha JK, Sambamoorthi U. Leading predictors of covid-19-related poor mental health in adult Asian Indians: An application of extreme gradient boosting and Shapley additive explanations. Int J Environ Res Public Health. 2022;20(1):775.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHussna AU, Trisha II, Ritun IJ, Alam MGR. Covid-19 impact on students' mental health: Explainable AI and classifiers. In: \u003cem\u003eInternational Conference on Decision Aid Sciences and Application (DASA): 2021; Sakheer, Bahrain\u003c/em\u003e: IEEE; 2021: 847\u0026ndash;851.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAdarsh V, Gangadharan GR. Applying explainable artificial intelligence models for understanding depression among IT workers. IT Prof. 2022;24(5):25\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoyce DW, Kormilitzin A, Smith KA, Cipriani A. Explainable artificial intelligence for mental health through transparency and interpretability for understandability. npj Digit Med. 2023;6(1):6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKotz J. Kalyakool Moort-Always Family: Strong culture, strong care, strong families: Codesigning a culturally considered approach to perinatal screening and support for the first 1000 days of parenthood from conception to infancy. Murdoch University; 2021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrinckley M-M, Calabria B, Walker J, Thurber KA, Lovett R. Reliability, validity, and clinical utility of a culturally modified Kessler scale (MK-K5) in the Aboriginal and Torres Strait Islander population. BMC Public Health. 2021;21(1):1111.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKwok SWH, Kotz J, Reibel T, Wang G, Walker R, Marriott R. Coupling machine learning models with an innovative technology-based screening tool for identifying psychological distress among Aboriginal perinatal mothers. In: \u003cem\u003e45th Annual International Conference of the IEEE Engineering in Medicine \u0026amp; Biology Society (EMBC)\u003c/em\u003e: 2023; Sydney: IEEE; 2023: 1\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKramer O. K-nearest neighbors. In: \u003cem\u003eDimensionality reduction with unsupervised nearest neighbors.\u003c/em\u003e Edited by Kramer O. Berlin, Heidelberg: Springer Berlin Heidelberg; 2013: 13\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuthaharan S. Support vector machine. In: \u003cem\u003eMachine learning models and algorithms for big data classification: Thinking with examples for effective learning.\u003c/em\u003e Edited by Suthaharan S. Boston, MA: Springer US; 2016: 207\u0026ndash;235.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMohamed ES, Naqishbandi TA, Bukhari SAC, Rauf I, Sawrikar V, Hussain A. A hybrid mental health prediction model using support vector machine, multilayer perceptron, and random forest algorithms. Healthc Analytics. 2023;3:100185.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCho G, Yim J, Choi Y, Ko J, Lee S-H. Review of machine learning algorithms for diagnosing mental illness. Psychiatry Invest. 2019;16(4):262.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhondoker M, Dobson R, Skirrow C, Simmons A, Stahl D. A comparison of machine learning methods for classification using simulation with multiple real data examples from mental health studies. Stat Methods Med Res. 2016;25(5):1804\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNori H, Jenkins S, Koch P, Caruana R. Interpretml: A unified framework for machine learning interpretability. arXiv preprint 2019:arXiv:1909.09223.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLou Y, Caruana R, Gehrke J, Hooker G. Accurate intelligible models with pairwise interactions. In: \u003cem\u003e19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining\u003c/em\u003e: 2013; 2013: 623\u0026ndash;631.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThimoteo LM, Vellasco MM, Amaral J, Figueiredo K, Yokoyama CL, Marques E. Explainable artificial intelligence for COVID-19 diagnosis through blood test variables. J Control Autom Electr Syst. 2022;33(2):625\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoldrack RA, Huckins G, Varoquaux G. Establishment of best practices for evidence for prediction: A review. JAMA Psychiatry. 2020;77(5):534\u0026ndash;40.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SM, Erion GG, Lee S-I. Consistent individualized feature attribution for tree ensembles. arXiv preprint 2018:arXiv:1802.03888.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRibeiro MT, Singh S, Guestrin C. Why should I trust you? Explaining the predictions of any classifier. In: \u003cem\u003e22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.\u003c/em\u003e San Francisco, California, USA: Association for Computing Machinery; 2016: 1135\u0026ndash;1144.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHong W, Lu Y, Zhou X, Jin S, Pan J, Lin Q, Yang S, Basharat Z, Zippi M, Goyal H. Usefulness of random forest algorithm in predicting severe acute pancreatitis. Front Cell Infect Microbiol. 2022;12:893294.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBelle V, Papantonis I. Principles and practice of explainable machine learning. Front Big Data. 2021;4:688969.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003e Government policies ensured Aboriginal children were systematically and forcibly removed from their families, communities and culture from the mid-1800s for almost a century. These children are referred to as the Stolen Generation or Stolen Children. These children were denied access to their culture and were punished if they spoke their language. The impacts of this are still being felt today (Healing Foundation, 2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Glassbox models allow users to trace the decision-making process back to individual data points or variables, making it easier to understand and interpret an algorithm's output.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Post-hoc agnostic models allow the decision-making mechanism to be a black box, and then develops a model that generates explanations after a recommendation has been made.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-4687233/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4687233/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003ePerinatal mental health significantly affects mothers, infants, and families. Despite their resilience and strengths, Aboriginal mothers experience disproportionate physical and mental health disparities. These result from historical and ongoing impacts of colonization and the resultant complex trauma. Conventional approaches to perinatal care present many barriers for Aboriginal mothers who frequently feel disengaged, apprehensive and unsafe. Current score-based risk-screening practices that algorithmically drive referrals, further ingrain fears including culturally biased judgments and child removal. The \u003cem\u003eBaby Coming You Ready\u003c/em\u003e (BCYR) model of care centred around a digitised, holistic, strengths-based assessment, was co-designed to address these barriers. The recent successful pilot demonstrated BCYR effectively replaced all current risk-based screens. However, many professionals disproportionately rely on psychological risk scores, overlooking the contextual circumstances of Aboriginal mothers, their cultural strengths and mitigating protective factors.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eTo address this singular reliance screening psychometrics whilst supporting strengthened culturally considered clinical assessment, we propose a culturally sensitive eXplainable AI (XAI) solution. It combines XAI with Aboriginal lived experience, knowledge and wisdom to generate a clinical prediction model to support professionals and Aboriginal mothers being screened. The XAI solution can identify, prioritise, and weigh both maternal protective strengths and risk factors, quantify their relative impacts on perinatal mental-health and well-being at both group and individual levels.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eDifferent machine learning algorithms, including Random Forest, K-nearest neighbour, and support vector machine, alongside glassbox Explainable Boosting Machine (EBM) models, were trained on the real life de-identified data generated during the BCYR pilot. Additionally, XAI techniques like SHAP and LIME are utilised for interpretability on black box models. Results show the EBM model demonstrates superior performance in prediction, with an accuracy of 0.849, F1 score of 0.771 and AUC of 0.821. Global explanations across the entire dataset and local explanations for individual cases, achieved through different methods, were compared and showed similar and stable results.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThis study demonstrated the potential for this XAI solution to enhance professionals' capability in culturally responsive clinical reasoning in perinatal mental-health screening to improve experience and strengthen outcomes for Aboriginal women.\u003c/p\u003e","manuscriptTitle":"Decoding Perinatal Mental Health: Investigating Protective and Risk Factors and Predictive Insights for Aboriginal Perinatal Mental Health through Explainable Machine Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-08-02 15:09:25","doi":"10.21203/rs.3.rs-4687233/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"96b5d2bc-4084-4103-8bd6-ad4af350f999","owner":[],"postedDate":"August 2nd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-09-16T04:48:34+00:00","versionOfRecord":[],"versionCreatedAt":"2024-08-02 15:09:25","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4687233","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4687233","identity":"rs-4687233","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}