Predicting Mental Health Treatment Seeking in the Technology Industry Using Machine Learning: A Comparative Analysis of Supervised Classification Models

Predicting Mental Health Treatment Seeking in the Technology Industry Using Machine Learning: A Comparative Analysis of Supervised Classification Models | 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 Predicting Mental Health Treatment Seeking in the Technology Industry Using Machine Learning: A Comparative Analysis of Supervised Classification Models Janakkumar Patel This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9257451/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 Mental health disorders affect approximately one in four people globally, yet treatment-seeking rates remain persistently low, particularly in high-stress professional environments such as the technology industry. Understanding the factors that predict whether an individual will seek mental health treatment is critical for designing effective workplace interventions. Methods This study applies five supervised machine learning classification algorithms — Random Forest, Logistic Regression, Support Vector Machine (SVM), Gaussian Naive Bayes, and eXtreme Gradient Boosting (XGBoost) — to predict treatment-seeking behaviour using the 2016 Open Sourcing Mental Illness (OSMI) survey dataset. The dataset includes responses from 1,434 technology industry workers across multiple countries. After preprocessing, including removal of high-missingness features and standardisation of categorical fields, a refined dataset of 960 entries was used for model training and evaluation. Feature correlation analysis was conducted to identify the strongest predictors of treatment-seeking behaviour. Results XGBoost achieved the highest classification accuracy of 88.7%, outperforming Random Forest (87.1%), Logistic Regression (87.1%), Gaussian Naive Bayes (86.6%), and SVM (85.6%). The most significant predictors of treatment-seeking behaviour were a prior diagnosis of a mental disorder and a family history of mental illness. A marked gender disparity was observed: male-identifying respondents reported substantially lower treatment-seeking rates despite similar rates of self-reported mental disorders. Conclusions Machine learning approaches, particularly XGBoost, demonstrate strong predictive capability for mental health treatment-seeking behaviour in technology industry workers. The identified gender disparity suggests a need for targeted workplace mental health interventions directed at male-identifying employees. These findings contribute to the growing evidence base for data-driven approaches to mental health decision support. mental health machine learning XGBoost treatment seeking technology industry classification workplace health predictive modelling Background Mental health conditions represent one of the most significant and underaddressed public health challenges of the twenty-first century. The World Health Organization estimates that approximately one in four people worldwide will be affected by a mental or neurological disorder at some point in their lives, yet the majority of those affected do not seek or receive professional treatment (Prizeman et al., 2023 ). This treatment gap is particularly pronounced in high-income, high-stress professional environments, where stigma, time constraints, and concerns about professional reputation create substantial barriers to help-seeking behaviour (Shatte et al., 2019 ). The technology industry presents a particularly important setting for the study of mental health treatment seeking. Technology workers are subject to distinctive stressors including rapid technological change, intense performance expectations, long working hours, and a workplace culture that has historically stigmatised mental health disclosure (Mitravinda et al., 2023 ). At the same time, technology companies typically offer comprehensive health benefit packages, meaning that treatment barriers are more likely to be attitudinal and social than financial. Understanding which factors most strongly predict whether a technology worker will seek treatment can inform the design of targeted interventions to reduce the treatment gap in this population. Machine learning approaches have emerged as powerful tools for identifying predictors of health-related behaviours from survey data. Supervised classification algorithms can model complex, non-linear relationships between demographic, occupational, and attitudinal variables and binary outcomes such as treatment seeking, often outperforming traditional logistic regression approaches when predictor sets are large and interactions are complex (Chung et al., 2022 ; Iyortsuun et al., 2023 ). XGBoost in particular has demonstrated strong performance in health prediction tasks due to its gradient boosting architecture, which sequentially improves predictions by correcting the errors of preceding models (Ali et al., 2023 ). The Open Sourcing Mental Illness (OSMI) survey provides a unique and publicly available dataset on mental health attitudes, experiences, and behaviours in the technology industry. The 2016 wave of this survey included over 1,400 responses from technology workers across multiple countries, capturing both personal factors such as age, gender, and family history of mental illness, and workplace factors such as employer-provided mental health benefits, remote work status, and perceived stigma around mental health disclosure. Several prior studies have used versions of this dataset to predict mental health outcomes (Katarya and Singhal, 2021 ; Khan and Dougherty, 2023 ; Baker et al., 2021 ), but comparative evaluation of multiple supervised classification models including XGBoost on the 2016 wave, with attention to gender-related disparities in treatment seeking, remains limited. This study addresses this gap by applying and systematically comparing five supervised classification algorithms to predict treatment-seeking behaviour in the 2016 OSMI dataset. The specific objectives are: (1) to identify the features most strongly correlated with treatment-seeking behaviour; (2) to evaluate and compare the performance of Random Forest, Logistic Regression, SVM, Gaussian Naive Bayes, and XGBoost classifiers; and (3) to characterise the observed gender disparity in treatment-seeking behaviour and discuss its implications for workplace mental health intervention design. Methods Dataset The 2016 OSMI Mental Health in Tech survey dataset was used for this analysis. The dataset is publicly available and contains 1,434 survey responses from technology industry workers across multiple countries. The survey included approximately 40 questions covering personal demographic characteristics — including age, gender identity, and country of residence — as well as workplace-specific questions regarding employer-provided mental health benefits, workplace attitudes toward mental illness, and the respondent's personal mental health history and treatment behaviour. The target variable for classification was the binary indicator of whether the respondent had sought professional mental health treatment. This variable was selected because it represents a clinically meaningful and actionable outcome, as distinct from diagnostic questions that require professional assessment. Data Preprocessing Feature Selection and Cleaning Initial inspection of the dataset revealed several features that were unsuitable for inclusion in classification models. Features containing primarily free-text responses were removed, as sentiment analysis or text classification would be required to encode them as model inputs, which was outside the scope of this comparative classification study. Features with high proportions of missing values (greater than 40 percent) were also removed. These preprocessing steps reduced the dataset from 1,434 to 960 complete entries suitable for model training and evaluation. Column names were standardised from full survey question text to abbreviated keyword labels to facilitate data manipulation and model interpretation. This refactoring preserved the semantic content of each feature while enabling efficient programmatic access. Handling of Age and Gender Anomalies The age feature contained anomalous values inconsistent with plausible respondent ages, including values below 18 and above 100. These entries were replaced with the dataset mean age of 34 years, consistent with standard practice for outlier imputation in survey data where the anomalous values are likely data entry errors rather than true observations (Chung et al., 2022 ). Gender identity was encoded into three categories — male, female, and genderqueer/other — based on the range of free-text responses provided. This categorisation was necessary to convert the open-ended gender field into a format suitable for supervised classification while respecting the diversity of gender identities represented in the dataset. Exploratory Data Analysis Prior to model training, exploratory data analysis was conducted to characterise the distribution of key variables and identify features most strongly associated with the target outcome. A Pearson correlation heatmap was computed for the top ten features most correlated with treatment-seeking behaviour. This analysis identified prior diagnosis of a mental disorder, history of a mental disorder, interference of mental health with work effectiveness when treated, and family history of mental illness as the four most strongly correlated predictors. Bivariate analysis of gender identity against both mental illness prevalence and treatment-seeking behaviour revealed a pronounced disparity. Male-identifying respondents reported similar rates of mental disorder to female-identifying respondents, but substantially lower rates of treatment seeking. This finding has important implications for workplace intervention design and is discussed further in the results and conclusions sections. Classification Models Five supervised classification algorithms were implemented and evaluated. All models were trained on the preprocessed dataset using stratified train/test splits to maintain class balance across splits. Accuracy, true positive rate (TPR), true negative rate (TNR), false positive rate (FPR), and false negative rate (FNR) were computed for each model. Random Forest Random Forest is an ensemble method that constructs multiple decision trees during training and outputs the class that is the mode of the classes of the individual trees (Rigatti, 2017 ). It is known for computational efficiency and resistance to overfitting relative to single decision trees. Each tree in the ensemble is trained on a bootstrap sample of the training data, and random feature subsets are used at each split, reducing variance and improving generalisation. Logistic Regression Logistic regression models the probability of the binary outcome as a sigmoid function of a linear combination of predictor variables (Hilbe, 2016 ). It is commonly used as a baseline comparator in binary classification tasks due to its interpretability and computational simplicity. Coefficients can be directly interpreted as log-odds ratios, facilitating clinical interpretation of predictor effects. Support Vector Machine Support Vector Machine (SVM) classification finds the hyperplane in feature space that maximally separates the two classes (Suthaharan, 2016 ). A radial basis function kernel was used to handle potential non-linear decision boundaries. SVM is known for strong performance in high-dimensional spaces but can be computationally demanding on large datasets. Gaussian Naive Bayes The Gaussian Naive Bayes classifier applies Bayes' theorem with the assumption that features are conditionally independent given the class label, and that continuous features follow a Gaussian distribution within each class (Cinar, 2023 ). While the independence assumption is rarely satisfied in practice, the model is computationally fast and often performs competitively on small to medium-sized datasets. XGBoost eXtreme Gradient Boosting (XGBoost) builds an ensemble of decision trees sequentially, with each tree trained to correct the residual errors of the preceding ensemble (Ali et al., 2023 ). Regularisation terms in the objective function control model complexity and reduce overfitting. XGBoost has demonstrated state-of-the-art performance on structured tabular data across a wide range of classification and regression tasks. Ethical Considerations This study used a publicly available, anonymised survey dataset collected with informed consent by the Open Sourcing Mental Illness organisation. No personally identifiable information was present in the dataset. As this study involved no direct contact with human participants and used only anonymised secondary data, formal ethical approval was not required. All analyses were conducted in accordance with the principles of data minimisation and respect for survey respondents' privacy. Results Feature Correlation Analysis The correlation analysis identified the following features as most strongly associated with treatment-seeking behaviour: prior diagnosis of a mental disorder (r = 0.73), history of a mental disorder in the past (r = 0.52), interference of untreated mental health with work effectiveness (r = 0.48), presence of a current mental disorder (r = 0.34), and family history of mental illness (r = 0.23). These findings are consistent with prior work showing that diagnostic status and family history are among the strongest predictors of help-seeking behaviour (Iyortsuun et al., 2023 ; Shatte et al., 2019 ). The correlation between mental health benefits provided by the employer and treatment seeking was positive but modest (r = 0.17), suggesting that benefit availability alone is insufficient to drive treatment uptake without accompanying cultural and attitudinal changes. The correlation between concern about career harm from mental health disclosure and treatment seeking was negative (r = -0.11), consistent with the hypothesis that stigma and career concerns act as barriers to help-seeking. Gender Disparity in Treatment Seeking Analysis of treatment-seeking behaviour stratified by gender identity revealed a substantial disparity. Female-identifying respondents showed notably higher rates of treatment seeking relative to their rates of self-reported mental disorder. Male-identifying respondents, despite reporting similar or higher absolute numbers of mental disorder diagnoses, showed substantially lower rates of treatment seeking. Genderqueer and non-binary respondents showed an intermediate pattern. This finding replicates and extends prior observations of gender-related help-seeking disparities in technology industry populations and is consistent with broader literature on male mental health stigma (Prizeman et al., 2023 ). Model Performance Comparison Table 1 presents the classification performance metrics for all five models. XGBoost achieved the highest accuracy at 88.7%, followed by Random Forest and Logistic Regression at 87.1%, Gaussian Naive Bayes at 86.6%, and SVM at 85.6%. All models performed within one standard deviation of each other, suggesting that the dataset is learnable across multiple model architectures without strong dependence on any specific algorithmic assumption. Table 1 . Classification performance metrics for all five supervised learning models. TPR = true positive rate; TNR = true negative rate; FPR = false positive rate; FNR = false negative rate. Model Accuracy TPR TNR FPR FNR Random Forest 87.1% 0.835 0.893 0.107 0.175 Logistic Regression 87.1% 0.789 0.919 0.081 0.211 SVM 85.6% 0.737 0.893 0.107 0.175 Gaussian Naive Bayes 86.6% 0.794 0.904 0.095 0.206 XGBoost 88.7% 0.854 0.902 0.098 0.145 XGBoost demonstrated the best balance between sensitivity (TPR = 0.854) and specificity (TNR = 0.902), indicating strong performance in correctly identifying both treatment-seeking and non-treatment-seeking respondents. Logistic Regression achieved the highest specificity (TNR = 0.919) but at the cost of lower sensitivity (TPR = 0.789), indicating a tendency to under-predict positive treatment-seeking outcomes. SVM achieved the lowest accuracy overall but remained within the competitive range. The relatively small performance differences between models — a range of only 3.1 percentage points in accuracy — suggests that the predictive signal in the dataset is strong and relatively insensitive to model choice. XGBoost's advantage likely derives from its ability to model non-linear interactions between predictors, particularly between diagnostic history, family history, and gender identity variables. Discussion This study demonstrates that supervised machine learning classification algorithms can predict mental health treatment-seeking behaviour in technology industry workers with accuracies approaching 89 percent using survey data alone. The strong predictive performance of all five models, and particularly XGBoost, aligns with a growing body of evidence supporting the use of machine learning in mental health informatics (Chung et al., 2022 ; Iyortsuun et al., 2023 ; Baba and Bunji, 2023 ). The identification of prior mental disorder diagnosis and family history of mental illness as the strongest predictors of treatment seeking is clinically intuitive and consistent with prior literature. Individuals who have previously been diagnosed are more likely to have established relationships with mental health professionals and reduced personal stigma around seeking care. Family history may operate through both genetic pathways — increasing baseline vulnerability to mental health conditions — and social learning pathways, where observing family members seek treatment normalises help-seeking behaviour. The gender disparity observed in this study represents one of the most practically significant findings. Male-identifying respondents reported disproportionately low treatment seeking relative to their rates of self-reported mental disorder. This pattern is consistent with broader epidemiological findings on male mental health stigma and the tendency for men to delay or avoid professional help for mental health concerns (Prizeman et al., 2023 ). In the technology industry, this pattern may be exacerbated by a workplace culture that valorises self-sufficiency and cognitive performance, making mental health struggles feel particularly incompatible with professional identity. Targeted interventions — including anonymous screening tools, male-focused mental health awareness campaigns, and leadership-level modelling of treatment-seeking behaviour — may be effective in reducing this disparity. The finding that employer-provided mental health benefits showed only modest correlation with treatment seeking has important implications for workplace policy. Simply providing benefit coverage may be insufficient if employees are unaware of available resources, perceive using those resources as risky to their career, or face logistical barriers such as appointment availability or time constraints. A more comprehensive approach combining benefit provision with proactive outreach, destigmatisation initiatives, and flexible access arrangements is likely to be more effective. Several limitations of this study should be acknowledged. The 2016 OSMI dataset, while unique in its focus on technology industry mental health, represents a single survey wave from a self-selected online respondent pool, which may not be fully representative of the global technology workforce. Respondents who chose to complete a mental health survey are likely to have greater awareness of and engagement with mental health issues than the broader population, which may limit the generalisability of findings to less engaged individuals. The binary treatment-seeking variable does not distinguish between types of treatment, frequency of treatment, or treatment outcomes, which are all important dimensions of help-seeking behaviour. Future work should seek to apply these models to more recent OSMI survey waves or comparable datasets to assess whether predictor relationships have changed over time, particularly given significant shifts in workplace mental health awareness since 2016. Incorporation of natural language processing on open-ended survey responses, explainable AI techniques such as SHAP values for feature attribution analysis, and multi-class prediction of specific treatment types rather than binary treatment-seeking would all represent meaningful extensions of the current analysis. Conclusions This study demonstrates that machine learning classification algorithms, particularly XGBoost, can predict mental health treatment-seeking behaviour among technology industry workers with high accuracy using survey-derived features. The most significant predictors are prior mental disorder diagnosis and family history of mental illness. A substantial gender disparity in treatment seeking was observed, with male-identifying respondents showing significantly lower help-seeking rates relative to their disorder prevalence. These findings have direct implications for the design of targeted workplace mental health interventions and support the broader application of machine learning to health decision support in occupational settings. Declarations Ethics approval and consent to participate This study used a publicly available, fully anonymised secondary dataset. No primary data collection was conducted and no direct contact with human participants took place. Formal ethics committee approval was not required for this study. Consent for publication Not applicable. Competing interests The author declares no competing interests. Use of Generative AI The author confirms that no generative AI or AI-assisted writing tools were used in the preparation of this manuscript. Funding This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Author Contribution JP conceptualised the study, conducted all data preprocessing and analysis, interpreted the results, and wrote the manuscript in its entirety. Acknowledgements The author thanks the Open Sourcing Mental Illness (OSMI) organisation for making the Mental Health in Tech survey dataset publicly available. Data Availability The dataset analysed in this study is publicly available from the Open Sourcing Mental Illness (OSMI) organisation and accessible via Kaggle at https://www.kaggle.com/datasets/osmi/mental-health-in-tech-survey. References Ali ZA, Abduljabbar ZH, Taher HA, Sallow AB, Almufti SM. Exploring the power of extreme gradient boosting algorithm in machine learning: A review. Acad J Nawroz Univ. 2023;12(2):320–34. Baba A, Bunji K. Prediction of mental health problem using annual student health survey: Machine learning approach. JMIR Mental Health. 2023;10:e42420. https://doi.org/10.2196/42420 . Baker JM, Cole C, Ross S. Survey-driven factors influencing mental health help-seeking in technology professionals. J Occup Health Psychol. 2021;26(3):215–28. Cinar A. Multi-class classification with the Gaussian Naive Bayes algorithm. J Data Appl. 2023;2:1–13. Chung YL, Mustapha M, Ibrahim A, Wahab NA. (2022). Mental health prediction using machine learning: Taxonomy, applications, and challenges. Applied Computational Intelligence and Soft Computing, 2022, 1–19. https://doi.org/10.1155/2022/9970363 Hilbe JM. Practical guide to logistic regression. CRC; 2016. Hill C, Waite P, Creswell C. Anxiety disorders in children and adolescents. Paediatrics Child Health. 2016;26(12):548–53. https://doi.org/10.1016/j.paed.2016.08.007 . Iyortsuun NK, Kim SH, Jhon M, Yang HJ, Pant S. A review of machine learning and deep learning approaches on mental health diagnosis. Healthcare. 2023;11(3):285. https://doi.org/10.3390/healthcare11030285 . Katarya R, Singhal S. (2021). Analysis of machine learning techniques for mental health prediction. Proceedings of the 2021 International Conference on Artificial Intelligence and Smart Systems, 73–79. IEEE. Khan T, Dougherty M. Predicting mental illness at workplace using machine learning. Mehran Univ Res J Eng Technol. 2023;42(1):95–108. Mitravinda A, Mitravinda KM, Nair DS, Srinivasa G. Mental health in tech: Analysis of workplace risk factors and impact of COVID-19. SN Comput Sci. 2023;4(2):197. https://doi.org/10.1007/s42979-022-01596-7 . Open Sourcing Mental Illness (OSMI). (2016). Mental health in tech survey 2016. Retrieved from https://www.kaggle.com/datasets/osmi/mental-health-in-tech-survey Piat M, Sabetti J, Couture A, Sylvestre J, Provencher H, Botschner J, Stayner D. What does recovery mean for me? Perspectives of Canadian mental health consumers. Psychiatr Rehabil J. 2009;32(3):199–207. https://doi.org/10.2975/32.3.2009.199.207 . Prizeman K, Weinstein N, McCabe C. Effects of mental health stigma on loneliness, social isolation, and relationships in young people with depression symptoms. BMC Psychiatry. 2023;23(1):527. https://doi.org/10.1186/s12888-023-05041-4 . Rigatti SJ. Random forest. J Insur Med. 2017;47(1):31–9. https://doi.org/10.17849/insm-47-01-31-39.1 . Shatte AB, Hutchinson DM, Teague SJ. Machine learning in mental health: A scoping review of methods and applications. Psychol Med. 2019;49(9):1426–48. https://doi.org/10.1017/S0033291719000151 . Suthaharan S. Support vector machine. Machine learning models and algorithms for big data classification. Springer; 2016. pp. 207–35. Vaishnavi K, Kamath UN, Rao BA, Reddy NS. (2022). Predicting mental health illness using machine learning algorithms. Journal of Physics: Conference Series, 2161(1), 012021. https://doi.org/10.1088/1742-6596/2161/1/012021 Wang X, Li H, Sun C, Zhang X, Wang T, Dong C, Guo D. Prediction of mental health in medical workers during COVID-19 based on machine learning. Front Public Health. 2021;9:697850. https://doi.org/10.3389/fpubh.2021.697850 . World Health Organization. World mental health report: Transforming mental health for all. WHO; 2022. 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-9257451","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":613970280,"identity":"a41ce048-9621-4846-a164-26fd82b7a919","order_by":0,"name":"Janakkumar Patel","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7UlEQVRIiWNgGAWjYBACxgYQPsAgx8/MfADIl5AhWouxZHtbAkgLD5E2HWBI3HDmjAGIQ1gLc/vZgx9nnLnD2HAj5/OrGzUWPAzsh49uwGtDT16y5IYbz5gZZ+Rus845BnQYT1raDfyOyjGQfPDhMBuzRO424xw2oBYJHjP8WvrfGP8EauFhk8h5ZpzzjxgtM3LMgA47LMHDc4b5cW4bUVremFnOOPPMQIK9zYw5t0+Ch42QXwz7c4xv9hy7U7//MPPjzznf6uT42Q8fw6+lAUwdABFsEmASn3IQkGdAaGH+QEj1KBgFo2AUjEwAAHZMUQ/GycJdAAAAAElFTkSuQmCC","orcid":"","institution":"Campbellsville University","correspondingAuthor":true,"prefix":"","firstName":"Janakkumar","middleName":"","lastName":"Patel","suffix":""}],"badges":[],"createdAt":"2026-03-29 08:38:54","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9257451/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9257451/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106425032,"identity":"e0ca6e9a-e8eb-4093-83c9-4d0eff389ed9","added_by":"auto","created_at":"2026-04-08 11:58:18","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":568574,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9257451/v1/cd40e184-ce2b-4d7b-b83b-4c9876609f3b.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Predicting Mental Health Treatment Seeking in the Technology Industry Using Machine Learning: A Comparative Analysis of Supervised Classification Models","fulltext":[{"header":"Background","content":"\u003cp\u003eMental health conditions represent one of the most significant and underaddressed public health challenges of the twenty-first century. The World Health Organization estimates that approximately one in four people worldwide will be affected by a mental or neurological disorder at some point in their lives, yet the majority of those affected do not seek or receive professional treatment (Prizeman et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This treatment gap is particularly pronounced in high-income, high-stress professional environments, where stigma, time constraints, and concerns about professional reputation create substantial barriers to help-seeking behaviour (Shatte et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe technology industry presents a particularly important setting for the study of mental health treatment seeking. Technology workers are subject to distinctive stressors including rapid technological change, intense performance expectations, long working hours, and a workplace culture that has historically stigmatised mental health disclosure (Mitravinda et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). At the same time, technology companies typically offer comprehensive health benefit packages, meaning that treatment barriers are more likely to be attitudinal and social than financial. Understanding which factors most strongly predict whether a technology worker will seek treatment can inform the design of targeted interventions to reduce the treatment gap in this population.\u003c/p\u003e \u003cp\u003eMachine learning approaches have emerged as powerful tools for identifying predictors of health-related behaviours from survey data. Supervised classification algorithms can model complex, non-linear relationships between demographic, occupational, and attitudinal variables and binary outcomes such as treatment seeking, often outperforming traditional logistic regression approaches when predictor sets are large and interactions are complex (Chung et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Iyortsuun et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). XGBoost in particular has demonstrated strong performance in health prediction tasks due to its gradient boosting architecture, which sequentially improves predictions by correcting the errors of preceding models (Ali et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe Open Sourcing Mental Illness (OSMI) survey provides a unique and publicly available dataset on mental health attitudes, experiences, and behaviours in the technology industry. The 2016 wave of this survey included over 1,400 responses from technology workers across multiple countries, capturing both personal factors such as age, gender, and family history of mental illness, and workplace factors such as employer-provided mental health benefits, remote work status, and perceived stigma around mental health disclosure. Several prior studies have used versions of this dataset to predict mental health outcomes (Katarya and Singhal, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Khan and Dougherty, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Baker et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), but comparative evaluation of multiple supervised classification models including XGBoost on the 2016 wave, with attention to gender-related disparities in treatment seeking, remains limited.\u003c/p\u003e \u003cp\u003eThis study addresses this gap by applying and systematically comparing five supervised classification algorithms to predict treatment-seeking behaviour in the 2016 OSMI dataset. The specific objectives are: (1) to identify the features most strongly correlated with treatment-seeking behaviour; (2) to evaluate and compare the performance of Random Forest, Logistic Regression, SVM, Gaussian Naive Bayes, and XGBoost classifiers; and (3) to characterise the observed gender disparity in treatment-seeking behaviour and discuss its implications for workplace mental health intervention design.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eDataset\u003c/h2\u003e \u003cp\u003eThe 2016 OSMI Mental Health in Tech survey dataset was used for this analysis. The dataset is publicly available and contains 1,434 survey responses from technology industry workers across multiple countries. The survey included approximately 40 questions covering personal demographic characteristics \u0026mdash; including age, gender identity, and country of residence \u0026mdash; as well as workplace-specific questions regarding employer-provided mental health benefits, workplace attitudes toward mental illness, and the respondent's personal mental health history and treatment behaviour.\u003c/p\u003e \u003cp\u003eThe target variable for classification was the binary indicator of whether the respondent had sought professional mental health treatment. This variable was selected because it represents a clinically meaningful and actionable outcome, as distinct from diagnostic questions that require professional assessment.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData Preprocessing\u003c/h3\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eFeature Selection and Cleaning\u003c/h2\u003e \u003cp\u003eInitial inspection of the dataset revealed several features that were unsuitable for inclusion in classification models. Features containing primarily free-text responses were removed, as sentiment analysis or text classification would be required to encode them as model inputs, which was outside the scope of this comparative classification study. Features with high proportions of missing values (greater than 40 percent) were also removed. These preprocessing steps reduced the dataset from 1,434 to 960 complete entries suitable for model training and evaluation.\u003c/p\u003e \u003cp\u003eColumn names were standardised from full survey question text to abbreviated keyword labels to facilitate data manipulation and model interpretation. This refactoring preserved the semantic content of each feature while enabling efficient programmatic access.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eHandling of Age and Gender Anomalies\u003c/h3\u003e\n\u003cp\u003eThe age feature contained anomalous values inconsistent with plausible respondent ages, including values below 18 and above 100. These entries were replaced with the dataset mean age of 34 years, consistent with standard practice for outlier imputation in survey data where the anomalous values are likely data entry errors rather than true observations (Chung et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eGender identity was encoded into three categories \u0026mdash; male, female, and genderqueer/other \u0026mdash; based on the range of free-text responses provided. This categorisation was necessary to convert the open-ended gender field into a format suitable for supervised classification while respecting the diversity of gender identities represented in the dataset.\u003c/p\u003e\n\u003ch3\u003eExploratory Data Analysis\u003c/h3\u003e\n\u003cp\u003ePrior to model training, exploratory data analysis was conducted to characterise the distribution of key variables and identify features most strongly associated with the target outcome. A Pearson correlation heatmap was computed for the top ten features most correlated with treatment-seeking behaviour. This analysis identified prior diagnosis of a mental disorder, history of a mental disorder, interference of mental health with work effectiveness when treated, and family history of mental illness as the four most strongly correlated predictors.\u003c/p\u003e \u003cp\u003eBivariate analysis of gender identity against both mental illness prevalence and treatment-seeking behaviour revealed a pronounced disparity. Male-identifying respondents reported similar rates of mental disorder to female-identifying respondents, but substantially lower rates of treatment seeking. This finding has important implications for workplace intervention design and is discussed further in the results and conclusions sections.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eClassification Models\u003c/h2\u003e \u003cp\u003eFive supervised classification algorithms were implemented and evaluated. All models were trained on the preprocessed dataset using stratified train/test splits to maintain class balance across splits. Accuracy, true positive rate (TPR), true negative rate (TNR), false positive rate (FPR), and false negative rate (FNR) were computed for each model.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eRandom Forest\u003c/h3\u003e\n\u003cp\u003eRandom Forest is an ensemble method that constructs multiple decision trees during training and outputs the class that is the mode of the classes of the individual trees (Rigatti, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). It is known for computational efficiency and resistance to overfitting relative to single decision trees. Each tree in the ensemble is trained on a bootstrap sample of the training data, and random feature subsets are used at each split, reducing variance and improving generalisation.\u003c/p\u003e\n\u003ch3\u003eLogistic Regression\u003c/h3\u003e\n\u003cp\u003eLogistic regression models the probability of the binary outcome as a sigmoid function of a linear combination of predictor variables (Hilbe, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). It is commonly used as a baseline comparator in binary classification tasks due to its interpretability and computational simplicity. Coefficients can be directly interpreted as log-odds ratios, facilitating clinical interpretation of predictor effects.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eSupport Vector Machine\u003c/h2\u003e \u003cp\u003eSupport Vector Machine (SVM) classification finds the hyperplane in feature space that maximally separates the two classes (Suthaharan, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). A radial basis function kernel was used to handle potential non-linear decision boundaries. SVM is known for strong performance in high-dimensional spaces but can be computationally demanding on large datasets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eGaussian Naive Bayes\u003c/h2\u003e \u003cp\u003eThe Gaussian Naive Bayes classifier applies Bayes' theorem with the assumption that features are conditionally independent given the class label, and that continuous features follow a Gaussian distribution within each class (Cinar, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). While the independence assumption is rarely satisfied in practice, the model is computationally fast and often performs competitively on small to medium-sized datasets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eXGBoost\u003c/h2\u003e \u003cp\u003eeXtreme Gradient Boosting (XGBoost) builds an ensemble of decision trees sequentially, with each tree trained to correct the residual errors of the preceding ensemble (Ali et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Regularisation terms in the objective function control model complexity and reduce overfitting. XGBoost has demonstrated state-of-the-art performance on structured tabular data across a wide range of classification and regression tasks.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eEthical Considerations\u003c/h2\u003e \u003cp\u003eThis study used a publicly available, anonymised survey dataset collected with informed consent by the Open Sourcing Mental Illness organisation. No personally identifiable information was present in the dataset. As this study involved no direct contact with human participants and used only anonymised secondary data, formal ethical approval was not required. All analyses were conducted in accordance with the principles of data minimisation and respect for survey respondents' privacy.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eFeature Correlation Analysis\u003c/h2\u003e \u003cp\u003eThe correlation analysis identified the following features as most strongly associated with treatment-seeking behaviour: prior diagnosis of a mental disorder (r\u0026thinsp;=\u0026thinsp;0.73), history of a mental disorder in the past (r\u0026thinsp;=\u0026thinsp;0.52), interference of untreated mental health with work effectiveness (r\u0026thinsp;=\u0026thinsp;0.48), presence of a current mental disorder (r\u0026thinsp;=\u0026thinsp;0.34), and family history of mental illness (r\u0026thinsp;=\u0026thinsp;0.23). These findings are consistent with prior work showing that diagnostic status and family history are among the strongest predictors of help-seeking behaviour (Iyortsuun et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Shatte et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe correlation between mental health benefits provided by the employer and treatment seeking was positive but modest (r\u0026thinsp;=\u0026thinsp;0.17), suggesting that benefit availability alone is insufficient to drive treatment uptake without accompanying cultural and attitudinal changes. The correlation between concern about career harm from mental health disclosure and treatment seeking was negative (r = -0.11), consistent with the hypothesis that stigma and career concerns act as barriers to help-seeking.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eGender Disparity in Treatment Seeking\u003c/h2\u003e \u003cp\u003eAnalysis of treatment-seeking behaviour stratified by gender identity revealed a substantial disparity. Female-identifying respondents showed notably higher rates of treatment seeking relative to their rates of self-reported mental disorder. Male-identifying respondents, despite reporting similar or higher absolute numbers of mental disorder diagnoses, showed substantially lower rates of treatment seeking. Genderqueer and non-binary respondents showed an intermediate pattern. This finding replicates and extends prior observations of gender-related help-seeking disparities in technology industry populations and is consistent with broader literature on male mental health stigma (Prizeman et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eModel Performance Comparison\u003c/h2\u003e \u003cp\u003eTable 1 presents the classification performance metrics for all five models. XGBoost achieved the highest accuracy at 88.7%, followed by Random Forest and Logistic Regression at 87.1%, Gaussian Naive Bayes at 86.6%, and SVM at 85.6%. All models performed within one standard deviation of each other, suggesting that the dataset is learnable across multiple model architectures without strong dependence on any specific algorithmic assumption.\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Classification performance metrics for all five supervised learning models. TPR\u0026thinsp;=\u0026thinsp;true positive rate; TNR\u0026thinsp;=\u0026thinsp;true negative rate; FPR\u0026thinsp;=\u0026thinsp;false positive rate; FNR\u0026thinsp;=\u0026thinsp;false negative rate.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003e\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" 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=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTPR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTNR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFPR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eFNR\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e87.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.835\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.893\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.107\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.175\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e87.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.789\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.919\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.081\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.211\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=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e85.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.737\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.893\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.107\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.175\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGaussian Naive Bayes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e86.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.794\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.904\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.095\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.206\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e88.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.854\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.902\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.098\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.145\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\u003eXGBoost demonstrated the best balance between sensitivity (TPR\u0026thinsp;=\u0026thinsp;0.854) and specificity (TNR\u0026thinsp;=\u0026thinsp;0.902), indicating strong performance in correctly identifying both treatment-seeking and non-treatment-seeking respondents. Logistic Regression achieved the highest specificity (TNR\u0026thinsp;=\u0026thinsp;0.919) but at the cost of lower sensitivity (TPR\u0026thinsp;=\u0026thinsp;0.789), indicating a tendency to under-predict positive treatment-seeking outcomes. SVM achieved the lowest accuracy overall but remained within the competitive range.\u003c/p\u003e \u003cp\u003eThe relatively small performance differences between models \u0026mdash; a range of only 3.1 percentage points in accuracy \u0026mdash; suggests that the predictive signal in the dataset is strong and relatively insensitive to model choice. XGBoost's advantage likely derives from its ability to model non-linear interactions between predictors, particularly between diagnostic history, family history, and gender identity variables.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study demonstrates that supervised machine learning classification algorithms can predict mental health treatment-seeking behaviour in technology industry workers with accuracies approaching 89 percent using survey data alone. The strong predictive performance of all five models, and particularly XGBoost, aligns with a growing body of evidence supporting the use of machine learning in mental health informatics (Chung et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Iyortsuun et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Baba and Bunji, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe identification of prior mental disorder diagnosis and family history of mental illness as the strongest predictors of treatment seeking is clinically intuitive and consistent with prior literature. Individuals who have previously been diagnosed are more likely to have established relationships with mental health professionals and reduced personal stigma around seeking care. Family history may operate through both genetic pathways \u0026mdash; increasing baseline vulnerability to mental health conditions \u0026mdash; and social learning pathways, where observing family members seek treatment normalises help-seeking behaviour.\u003c/p\u003e \u003cp\u003eThe gender disparity observed in this study represents one of the most practically significant findings. Male-identifying respondents reported disproportionately low treatment seeking relative to their rates of self-reported mental disorder. This pattern is consistent with broader epidemiological findings on male mental health stigma and the tendency for men to delay or avoid professional help for mental health concerns (Prizeman et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In the technology industry, this pattern may be exacerbated by a workplace culture that valorises self-sufficiency and cognitive performance, making mental health struggles feel particularly incompatible with professional identity. Targeted interventions \u0026mdash; including anonymous screening tools, male-focused mental health awareness campaigns, and leadership-level modelling of treatment-seeking behaviour \u0026mdash; may be effective in reducing this disparity.\u003c/p\u003e \u003cp\u003eThe finding that employer-provided mental health benefits showed only modest correlation with treatment seeking has important implications for workplace policy. Simply providing benefit coverage may be insufficient if employees are unaware of available resources, perceive using those resources as risky to their career, or face logistical barriers such as appointment availability or time constraints. A more comprehensive approach combining benefit provision with proactive outreach, destigmatisation initiatives, and flexible access arrangements is likely to be more effective.\u003c/p\u003e \u003cp\u003eSeveral limitations of this study should be acknowledged. The 2016 OSMI dataset, while unique in its focus on technology industry mental health, represents a single survey wave from a self-selected online respondent pool, which may not be fully representative of the global technology workforce. Respondents who chose to complete a mental health survey are likely to have greater awareness of and engagement with mental health issues than the broader population, which may limit the generalisability of findings to less engaged individuals. The binary treatment-seeking variable does not distinguish between types of treatment, frequency of treatment, or treatment outcomes, which are all important dimensions of help-seeking behaviour.\u003c/p\u003e \u003cp\u003eFuture work should seek to apply these models to more recent OSMI survey waves or comparable datasets to assess whether predictor relationships have changed over time, particularly given significant shifts in workplace mental health awareness since 2016. Incorporation of natural language processing on open-ended survey responses, explainable AI techniques such as SHAP values for feature attribution analysis, and multi-class prediction of specific treatment types rather than binary treatment-seeking would all represent meaningful extensions of the current analysis.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThis study demonstrates that machine learning classification algorithms, particularly XGBoost, can predict mental health treatment-seeking behaviour among technology industry workers with high accuracy using survey-derived features. The most significant predictors are prior mental disorder diagnosis and family history of mental illness. A substantial gender disparity in treatment seeking was observed, with male-identifying respondents showing significantly lower help-seeking rates relative to their disorder prevalence. These findings have direct implications for the design of targeted workplace mental health interventions and support the broader application of machine learning to health decision support in occupational settings.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e \u003cp\u003eThis study used a publicly available, fully anonymised secondary dataset. No primary data collection was conducted and no direct contact with human participants took place. Formal ethics committee approval was not required for this study.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe author declares no competing interests.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eUse of Generative AI\u003c/h2\u003e \u003cp\u003eThe author confirms that no generative AI or AI-assisted writing tools were used in the preparation of this manuscript.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eJP conceptualised the study, conducted all data preprocessing and analysis, interpreted the results, and wrote the manuscript in its entirety.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eThe author thanks the Open Sourcing Mental Illness (OSMI) organisation for making the Mental Health in Tech survey dataset publicly available.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe dataset analysed in this study is publicly available from the Open Sourcing Mental Illness (OSMI) organisation and accessible via Kaggle at https://www.kaggle.com/datasets/osmi/mental-health-in-tech-survey.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAli ZA, Abduljabbar ZH, Taher HA, Sallow AB, Almufti SM. Exploring the power of extreme gradient boosting algorithm in machine learning: A review. Acad J Nawroz Univ. 2023;12(2):320\u0026ndash;34.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaba A, Bunji K. Prediction of mental health problem using annual student health survey: Machine learning approach. JMIR Mental Health. 2023;10:e42420. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/42420\u003c/span\u003e\u003cspan address=\"10.2196/42420\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaker JM, Cole C, Ross S. Survey-driven factors influencing mental health help-seeking in technology professionals. J Occup Health Psychol. 2021;26(3):215\u0026ndash;28.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCinar A. Multi-class classification with the Gaussian Naive Bayes algorithm. J Data Appl. 2023;2:1\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChung YL, Mustapha M, Ibrahim A, Wahab NA. (2022). Mental health prediction using machine learning: Taxonomy, applications, and challenges. Applied Computational Intelligence and Soft Computing, 2022, 1\u0026ndash;19. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2022/9970363\u003c/span\u003e\u003cspan address=\"10.1155/2022/9970363\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHilbe JM. Practical guide to logistic regression. CRC; 2016.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHill C, Waite P, Creswell C. Anxiety disorders in children and adolescents. Paediatrics Child Health. 2016;26(12):548\u0026ndash;53. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.paed.2016.08.007\u003c/span\u003e\u003cspan address=\"10.1016/j.paed.2016.08.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIyortsuun NK, Kim SH, Jhon M, Yang HJ, Pant S. A review of machine learning and deep learning approaches on mental health diagnosis. Healthcare. 2023;11(3):285. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/healthcare11030285\u003c/span\u003e\u003cspan address=\"10.3390/healthcare11030285\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKatarya R, Singhal S. (2021). Analysis of machine learning techniques for mental health prediction. Proceedings of the 2021 International Conference on Artificial Intelligence and Smart Systems, 73\u0026ndash;79. IEEE.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhan T, Dougherty M. Predicting mental illness at workplace using machine learning. Mehran Univ Res J Eng Technol. 2023;42(1):95\u0026ndash;108.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMitravinda A, Mitravinda KM, Nair DS, Srinivasa G. Mental health in tech: Analysis of workplace risk factors and impact of COVID-19. SN Comput Sci. 2023;4(2):197. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s42979-022-01596-7\u003c/span\u003e\u003cspan address=\"10.1007/s42979-022-01596-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOpen Sourcing Mental Illness (OSMI). (2016). Mental health in tech survey 2016. Retrieved from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.kaggle.com/datasets/osmi/mental-health-in-tech-survey\u003c/span\u003e\u003cspan address=\"https://www.kaggle.com/datasets/osmi/mental-health-in-tech-survey\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePiat M, Sabetti J, Couture A, Sylvestre J, Provencher H, Botschner J, Stayner D. What does recovery mean for me? Perspectives of Canadian mental health consumers. Psychiatr Rehabil J. 2009;32(3):199\u0026ndash;207. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2975/32.3.2009.199.207\u003c/span\u003e\u003cspan address=\"10.2975/32.3.2009.199.207\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePrizeman K, Weinstein N, McCabe C. Effects of mental health stigma on loneliness, social isolation, and relationships in young people with depression symptoms. BMC Psychiatry. 2023;23(1):527. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12888-023-05041-4\u003c/span\u003e\u003cspan address=\"10.1186/s12888-023-05041-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRigatti SJ. Random forest. J Insur Med. 2017;47(1):31\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.17849/insm-47-01-31-39.1\u003c/span\u003e\u003cspan address=\"10.17849/insm-47-01-31-39.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShatte AB, Hutchinson DM, Teague SJ. Machine learning in mental health: A scoping review of methods and applications. Psychol Med. 2019;49(9):1426\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/S0033291719000151\u003c/span\u003e\u003cspan address=\"10.1017/S0033291719000151\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuthaharan S. Support vector machine. Machine learning models and algorithms for big data classification. Springer; 2016. pp. 207\u0026ndash;35.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVaishnavi K, Kamath UN, Rao BA, Reddy NS. (2022). Predicting mental health illness using machine learning algorithms. Journal of Physics: Conference Series, 2161(1), 012021. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1088/1742-6596/2161/1/012021\u003c/span\u003e\u003cspan address=\"10.1088/1742-6596/2161/1/012021\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang X, Li H, Sun C, Zhang X, Wang T, Dong C, Guo D. Prediction of mental health in medical workers during COVID-19 based on machine learning. Front Public Health. 2021;9:697850. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpubh.2021.697850\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2021.697850\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organization. World mental health report: Transforming mental health for all. WHO; 2022.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"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":"mental health, machine learning, XGBoost, treatment seeking, technology industry, classification, workplace health, predictive modelling","lastPublishedDoi":"10.21203/rs.3.rs-9257451/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9257451/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eMental health disorders affect approximately one in four people globally, yet treatment-seeking rates remain persistently low, particularly in high-stress professional environments such as the technology industry. Understanding the factors that predict whether an individual will seek mental health treatment is critical for designing effective workplace interventions.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis study applies five supervised machine learning classification algorithms \u0026mdash; Random Forest, Logistic Regression, Support Vector Machine (SVM), Gaussian Naive Bayes, and eXtreme Gradient Boosting (XGBoost) \u0026mdash; to predict treatment-seeking behaviour using the 2016 Open Sourcing Mental Illness (OSMI) survey dataset. The dataset includes responses from 1,434 technology industry workers across multiple countries. After preprocessing, including removal of high-missingness features and standardisation of categorical fields, a refined dataset of 960 entries was used for model training and evaluation. Feature correlation analysis was conducted to identify the strongest predictors of treatment-seeking behaviour.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eXGBoost achieved the highest classification accuracy of 88.7%, outperforming Random Forest (87.1%), Logistic Regression (87.1%), Gaussian Naive Bayes (86.6%), and SVM (85.6%). The most significant predictors of treatment-seeking behaviour were a prior diagnosis of a mental disorder and a family history of mental illness. A marked gender disparity was observed: male-identifying respondents reported substantially lower treatment-seeking rates despite similar rates of self-reported mental disorders.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eMachine learning approaches, particularly XGBoost, demonstrate strong predictive capability for mental health treatment-seeking behaviour in technology industry workers. The identified gender disparity suggests a need for targeted workplace mental health interventions directed at male-identifying employees. These findings contribute to the growing evidence base for data-driven approaches to mental health decision support.\u003c/p\u003e","manuscriptTitle":"Predicting Mental Health Treatment Seeking in the Technology Industry Using Machine Learning: A Comparative Analysis of Supervised Classification Models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-31 08:42:00","doi":"10.21203/rs.3.rs-9257451/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":"59474163-60f1-4ce3-b25c-a2bbee1e0d1e","owner":[],"postedDate":"March 31st, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-08T11:57:34+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-31 08:42:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9257451","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9257451","identity":"rs-9257451","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}