Rural-Urban Disparities in Early Childhood Development: Insights from Interpretable Ensemble Machine Learning Model

Rural-Urban Disparities in Early Childhood Development: Insights from Interpretable Ensemble Machine Learning Model | 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 Rural-Urban Disparities in Early Childhood Development: Insights from Interpretable Ensemble Machine Learning Model Obaloluwa David GBADEGESIN, Babatunde Makinde Gbadebo, Joshua Odunayo AKINYEMI This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8615011/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Background Early childhood development (ECD) is essential for lifelong health, education, and economic productivity, as emphasized by SDG Target 4.2, which seeks universal access to quality pre-primary education by 2030. In Nigeria, ECD are significantly impacted by poverty, malnutrition, and systemic inequities. However, there is limited research on context-specific determinants such as caregiving practices, parental discipline, and home environments. This study developed a Multilayer Stacked Weighted Ensemble Model (MSWEM) and analyse data from the 2021 Nigeria Multiple Indicator Cluster Survey (MICS), comprising 15,899 children aged 24–59 months. Statistical techniques included descriptive analysis, binary logistic regression for predictor identification and machine learning for advanced prediction and SHAP (SHapley Additive exPlanations) analysis for model interpretability. The model development was implemented in Python 3.7, integrating ensemble methods to enhance prediction accuracy Results The MSWEM achieved a high predictive accuracy with an ROC-AUC of 93% and accuracy of 85.2%. The model correctly predicted 46.7% on – track children in urban areas and 24.7% on-track children in rural areas. Factors influencing the likelihood of a child being on track include age of the child, cognitive caregiving practice, overall physical home environment, mother’s education, age of woman, to mention but a few Conclusion This study highlights the utility of interpretable machine learning in understanding and predicting ECD outcomes. These efforts are essential for ensuring equitable opportunities for all children in Nigeria, aligning with national development goals and global progress toward SDG 4.2. Early childhood development interpretable ensemble models machine learning caregiving practices SDG Target 4.2. Count: 244 words Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 BACKGROUND Early childhood development (ECD) is a cornerstone of human potential, laying the groundwork for lifelong health, education, and economic success. It primarily focuses on children aged eight years and younger, encompassing the development of motor, language, cognitive, socio-emotional, and self-regulatory skills (UNICEF, 2023). These abilities develop during the early years of life, a unique and critical period characterized by extraordinary brain plasticity—the capacity of the brain to adapt and grow in response to experiences. This heightened plasticity offers both immense opportunities for growth and profound vulnerabilities, as the brain’s development is heavily influenced by early experiences (Spencer et al., 2019 ). Research demonstrates that timely support during this critical period yields far greater benefits than efforts made later in life to compensate for early disadvantages (Britto et al., 2017 ). For this reason, ECD interventions are vital for fostering human development and breaking intergenerational cycles of poverty. Despite significant strides in child survival, such as a 53% reduction in under-five mortality between 1990 and 2015, the burden of developmental risk remains high. Approximately 250 million children under five years of age in low- and middle-income countries (LMICs) are at risk of not reaching their full developmental potential, a figure that rises to two-thirds in Sub-Saharan Africa (Black et al., 2016 ). These risks are compounded by factors such as extreme poverty and stunting, which not only hinder individual potential but also impose significant societal costs. For instance, the mean annual loss in adult income due to these risks is estimated at 26%, perpetuating poverty and stunting economic growth (Richter et al., 2017). Eliminating stunting alone could have increased per capita GDP by 7%, according to World Bank estimates (Galasso & Wagstaff, 2018 ). The risks extend beyond poverty and malnutrition to include factors like inadequate caregiving practices, parental discipline and violence, the physical home environment, and child functioning difficulties. These cumulative risks threaten not just individual developmental outcomes but also the human capital that drives societal progress. As such, ECD forms the foundation of human capital, accounting for a significant portion of the wealth disparities between nations (Lange et al., 2018). Existing research on ECD in Nigeria often focuses on factors such as poverty and malnutrition while context-specific determinants like caregiving practices, parental discipline, exposure to violence, and the physical home environment are scarcely investigated. This limited scope fails to capture the complex interplay of factors influencing child development and constrains efforts to design effective, evidence-based interventions tailored to the Nigerian context. Furthermore, public health and social systems in Nigeria are often unable to identify children at risk of developmental delays early for prompt and effective interventions. Traditional approaches to analysis of ECD risk factors have relied on linear models that emphasize individual predictors but fail to account for non-linear relationships and interactions among variables. While machine learning (ML) offers advanced predictive capabilities and the potential to uncover hidden patterns in large datasets, its application in ECD research in Nigeria remains minimal. The "black box" nature of many ML models further complicates their adoption in public health, where interpretability is essential for translating findings into actionable policies. Emerging approaches such as explainable AI (XAI) address this challenge, offering tools to enhance the interpretability of complex ML models. This study aims to bridge these gaps by leveraging interpretable ensemble machine learning models for early childhood development and analyse the socio-demographic, environmental, and caregiving factors influencing ECD outcomes. The strong alignment of this study’s findings with Bronfenbrenner’s PPCT Bioecological Theory of Human Development is also evident. By incorporating underexplored determinants, such as parental caregiving practices, home environments, and exposure to violence, this research will provide a more comprehensive understanding of ECD disparities in Nigeria. The findings will inform the development of targeted interventions and policies to address these disparities, reduce developmental inequities, and enhance developmental outcomes for Nigerian children. METHODS DATA SOURCE AND DESCRIPTION This study utilizes data from the sixth round of the Nigeria Multiple Indicator Cluster Survey conducted in 2021. The study population consists of children aged 2 to 5 years and their families, drawn from the nationally representative sample of the 2021 MICS 6 dataset for Nigeria. The data was collected through a multi-stage stratified cluster sampling approach. This age cohort was selected to align with the study’s objective of examining early childhood development and predictors. STUDY VARIABLES The new measure of early childhood development which is the Early Childhood Development Index 2030 was estimated from a list of 20 questions asked in the 2021 MICS6 round according to age groups and this was used to generate a single summative score (United Nations Children’s Fund, 2023 ). ECDI2030 is given as the proportion of children aged 24 to 59 months who are developmentally on track. The dependent variable was thus obtained by generating a variable that identifies children with on-track development according to the pre-defined age-specific cut-scores based on total milestones achieved from the 20 items (UNICEF, 2023).This variable generated was used to report the proportion of children aged 24 to 59 months who were developmentally on-track in health, learning and psychosocial wellbeing. Independent variables included age of child in month, gender, age of mother/caregiver, education, wealth index, zone, household characteristics, caregiving practices, parental discipline and violence practices, child functioning difficulties. Independent variables were selected based on existing literature on early childhood development (Bornstein et al., 2022 ) DATA MANAGEMENT AND ANALYSIS Data Preprocessing The MICS data was extensively pre-processed to ensure that it is suitable for machine learning analysis. To address missing data in this study, a multiple imputation technique using chained equations was used, following the approach described by Azur et al. ( 2011 ). This method involves running several regression models to predict missing values based on other observed variables in the dataset. The type of regression model used depends on the nature of the variable being imputed. Likewise, class imbalance was observed in the target variable -whether a child is developmentally on-track or off-track, and this was handled using the Synthetic Minority Oversampling Technique (SMOTE). SMOTE is an effective oversampling method that generates synthetic data points for the minority class, helping to create a balanced dataset and improve model performance. MODEL DEVELOPMENT This study utilized advanced ensemble machine learning techniques, including stacking and bagging, to develop predictive models with superior performance and reduced variance. Stack ensembling involves training multiple base models and combining their predictions using a meta-model (stacker). Multi-layer stacking extends this by creating hierarchical layers of stackers, where each layer builds upon the predictions of the previous one. In this approach, we have base models and stacker models and AutoGluon’s framework was used to introduce novel enhancements such as reusing base model architectures, incorporation of original features, ensemble selection in the final layer using a greedy weighted algorithm as highlighted by Erickson et al., (2020). The dataset was partitioned into three subsets comprising 80% for training, 10% for validation, and 10% for testing. To optimize the use of training data and reduce prediction variance, k -fold bagging was employed across all layers of the ensemble architecture, in line with the approach described by Erickson et al. (2020). This method involves dividing the training data into k equal folds, sequentially training the model on k–1 folds and validating on the remaining fold. Through this process, every data point is eventually utilized for both training and validation, thereby improving the model's generalization capacity and minimizing overfitting. To further stabilize predictions, repeated k -fold bagging was implemented. Overall, the total number of models trained within the ensemble is given by the expression M × N × K + 1 , where M denotes the number of stacking layers (excluding the final aggregation layer), N is the number of model types per layer, K is the number of bagging folds, and the additional 1 corresponds to the final meta-model trained using a Greedy Weighted Averaging algorithm. The resulting ensemble model was subsequently used to generate predictions for both rural and urban populations. According to Lundberg and Lee (2018), the best explanation for a simple model is the model itself, as it inherently represents its decision process in a straightforward manner. However, for complex models, which lack inherent interpretability, an explanation model must be used. This explanation model g is defined as an interpretable approximation of the original prediction model f , focusing on local explanations based on a specific input x . Kernel SHAP was employed in this study to achieve model interpretability. The performance of the classification models was assessed using standard evaluation metrics, including Accuracy, Precision, Recall (Sensitivity), F1 Score, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and the Area Under the Receiver Operating Characteristic Curve (ROC_AUC). RESULTS Sociodemographic characteristics of participants Data from 15,899 households/children were analysed by filtering the MICS dataset to include children aged 24 to 59 months. Of these, the average age of the children included in the study was 41.3 months (SD = 10.3), with a relatively even distribution by gender: 49.5% male and 50.4% female. The sociodemographic characteristics of primary caregivers examined in this study were age and educational status as shown in Table 1. On average, primary caregivers were 31.2 years old (SD = 7.1). Also, Table 1 also revealed that the educational backgrounds of primary caregivers varied with 42.3% reported having no formal education, 15.9% completed only primary education, 6.7% attained junior secondary school education while 25.1% completed senior secondary education, and 10.0% received higher/tertiary education. In terms of household characteristics, the average household size in the sample was 7.9 individuals (SD = 4.0). On average, households had 4.95 children under the age of 18 years (SD = 2.98), with 2.11 children under the age of five years (SD = 1.1). Additionally, a newly created variable estimating household crowding indicated that, on average, 3.39 people shared a bedroom (SD = 1.47). On average, families owned 3.57 (SD = 1.38) out of five possible electronic assets, 1.88 (SD = 1.11) out of seven possible family amenities, and 1.92 (SD = 1.04) out of five possible learning materials. Regarding housing quality, 31.5% of families used unimproved water sources, and 24.5% lacked adequate or used unimproved toileting facilities. Additionally, 34.7% of homes had natural floors, 13.4% had natural roofs, and 35.4% had natural exterior walls. A significant majority (96.1%) of families cooked with solid, less healthy fuel sources, and 45.5% cooked outside the home. Table 1 : Background Characteristics of All Participants Mean (SD) Frequency (%) Child characteristics N = 15899 Mean age of child 41.3 (10.3) N/A Gender Male N/A 49.5 Female N/A 50.4 Primary Caregivers Mean Age 31.2 (7.1) Education No formal N/A 42.3 Primary N/A 15.9 Junior Secondary N/A 6.7 Senior Secondary N/A 25.1 Higher / Tertiary N/A 10.0 Household characteristics Household size 7.90 (4.0) N/A Number of children under 18 years in an household 4.95 (2.98) N/A Number of children under 5 years in an household 2.11 (1.1) N/A 3.39 (1.47) N/A Early Childhood Development The new measure of early childhood development which is the Early Childhood Development Index 2030 was estimated from a list of 20 question asked in the 2021 MICS6 round according to age groups. The total milestones according to specific age groups was used to classify under five children as either developmentally on-track or off-track according to the ECDI2030 Technical Manual (UNICEF, 2023). The result revealed that in a sample of 15899 children in Nigeria, the total on-track children is 38.2% . The result was further disaggregated by sex, area, mother’s education, functional difficulties, wealth index and geopolitical zone as presented in Table 2. Table 2: Percentage of Children aged 2 – 4 years who are developmentally on track in health, learning and psychosocial well-being in Nigeria Percentage of Children who are developmentally on-track Number of children aged 2 – 4 years Total 38.2% 15899 Sex Female Male 39.1% 37.3% 7904 8039 Area Rural Urban 29.2% 55.8% 10552 5391 Mother’s education None Primary Junior Secondary Senior Secondary Higher / tertiary 22.7% 37.5% 41.5% 51.9% 69.0% 6749 2531 1071 3995 1598 Functional disabilities Has functional disabilities Has no functional disabilities 14.3% 39.3% 671 15272 Wealth index quintile Poorest Richest Second Middle Fourth 22.2% 68.0% 27.8% 37.0% 52.8% 4138 2335 3698 3098 2675 Geopolitical zone North West North East North Central South South South East South West 27.0% 26.9% 34.2% 53.1% 53.2% 65.6% 5651 2811 2197 1657 1545 2082 EVALUATION OF THE MULTILAYER STACKED WEIGHTED ENSEMBLE MODEL ON THE TRAINING SET, VALIDATION SET AND TEST SET. Figure 1 shows the model architecture of a stacked ensemble model, highlighting the steps taken to achieve optimal predictive accuracy. This model was developed using multiple layers of base learners, each contributing to the final ensemble prediction. The performance of the Multilayer Stacked Weighted Ensemble Model was evaluated using various metrics on the training, validation, and test datasets. The model achieved an accuracy of 100% on the training set, demonstrating its ability to correctly classify nearly all cases in the training data. On the validation set, accuracy became 83.7%, and on the test set, it increased to 85.2%. Also, the assessment of the model predicting power was compared using the validation and test set and this is shown in Figure 2. Performances were optimum in both validation and test set. Classification Performance Table 3 shows the predictive performance of the newly developed model across rural, urban areas in Nigeria. A total of 24.7% were predicted to be on-track in their early childhood development in rural areas while 46.4% of children were predicted to be on-track in urban areas in Nigeria. Table 3: Performance of MSWEM in classifying on-track and off-track in rural area and urban areas. Rural Sensitivity Specificity F1-score Off track (0) 98% 91% 94% On track (1) 77% 93% 84% Accuracy = 92% ROC_AUC = 87% Urban Off track (0) 92% 88% 90% On track (1) 87% 91% 89% Accuracy = 90% ROC_AUC = 90% MODEL INTERPRETATION IN RURAL AND URBAN AREAS Figure 3 presents SHAP summary plots comparing feature importance in rural and urban areas. In both contexts, a child’s age is the most influential factor. In rural areas, maternal age plays a key role, with older mothers more likely to have children on track developmentally. In contrast, maternal education was the second most important factor in urban areas, where higher educational attainment strongly correlates with better child development. Although maternal education influences both contexts, it ranks lower in rural areas, reflecting lower educational attainment among rural mothers. The wealth index shows a stark contrast, ranking highly in urban areas but significantly lower in rural regions, indicating wealth concentration in cities. Interaction effects within urban areas further highlight the complex relationship between wealth and child development. The home environment is a critical predictor in both settings, though its variability is lower in rural areas. In urban areas, better home environments strongly correlate with improved child outcomes. Cognitive caregiving exhibits less variability in rural areas but remains a strong positive influence across both settings. Similarly, socioemotional caregiving ranks lower in both areas, yet SHAP plots reveal interactions underscoring its importance in shaping developmental outcomes. INTERACTION OF VARIABLES In rural areas, a negative relationship between age and SHAP values indicates that younger children (25–35 months) as shown in figure 4 benefit more from age-related developmental gains, but this effect diminishes as they grow older. Functional difficulties consistently lower SHAP values, reducing development across all ages. Household wealth and geographic zone have a stronger influence on children under five, while older children from wealthier families and favourable zones (South-South and South-West) are more likely to be on track. Older mothers engage more in socioemotional activities, and higher cognitive care strongly enhances developmental outcomes, especially in older children. Socioemotional care has a greater impact on younger children but varies across age groups. Additional interactions exist between physical violence and child age, home environment and wealth index, and household crowding and child age. Conversely, SHAP dependence plots for children in urban areas shown in Figure 5 reveal that maternal education is a significant predictor of early childhood development (ECD) in urban areas. Interaction effects between maternal education and the child’s age were observed. The analysis indicates that higher SHAP values for maternal education were associated with children whose mothers possess higher levels of education. Higher SHAP values, which positively influence predictions of children being on track, correlate with increased levels of maternal education and the number of cognitive activities to which a child is exposed. Conversely, lower levels of exposure to cognitive caregiving were associated with decreased SHAP values, while higher exposure to socioemotional caregiving practices mitigated this effect. Furthermore, the analysis reveals that a smaller household size increases the SHAP values for the physical environment, thereby enhancing predictions of a child being on track. These findings underscore the critical role of maternal education, cognitive stimulation, and household composition in shaping developmental outcomes in urban contexts. INTERPRETATION OF THE MODEL IN RURAL AND URBAN AREAS Following an examination of variable interactions, it became evident that further investigation into the contributions of each variable to predicting early childhood development (ECD) in rural and urban areas was necessary. An individualized explanation of how variables influence ECD predictions in these contexts is presented through the SHAP force plots in Figures 6 and 7. These plots provide detailed insights into model decision-making for rural and urban areas, respectively. In Figure 6, which focuses on the rural area, the SHAP force plot illustrates how various features contribute to the prediction for a single child. Features highlighted in red exert a positive influence on the final prediction, while those in blue negatively impact the outcome. The force plot represents a specific data point where the model predicts whether the child is on-track or off-track. RURAL AREA The feature "Crowd" exhibits a SHAP value of 4.5, significantly elevating the prediction above the base value. Positive influences on the prediction include the child’s exposure to non-violent disciplinary practices, psychological aggression activities, the physical home environment, and the child's age, all of which contribute to an increased likelihood of being on track. Conversely, factors that lowered the prediction include a large household size and the young age of the mother. In this instance, the prediction for the child was off-track, primarily driven by the negative impact of larger family sizes, residence in the Northeast zone (represented as 2), and the young maternal age. For children predicted to be on-track, the force plot reveals positive driving factors that elevated the prediction beyond the base value. These include exposure to two cognitive caregiving activities, residence in a wealthier household, engagement in two psychological aggression activities, the quality of the home environment, as well as the age of the mother and child. This individualized analysis underscores the intricate balance between positive and negative factors influencing ECD outcomes in rural contexts. URBAN AREA Figure 7a presents an analysis of an off-track prediction for a child residing in an urban area. The SHAP force plot highlights the variables that contribute positively and negatively to the final prediction. Factors that push the prediction higher relative to the base value include the quality of the physical home environment, the presence of four household members, a middle wealth index quintile, residence in the Northeast zone, and a 23-year-old mother with secondary education who engages the child in two cognitive caregiving activities. Conversely, variables that lowered the prediction include the occurrence of physical violence, the child’s age (50 months), and exposure to only one socioemotional caregiving activity. These factors collectively diminish the likelihood of the child being on track. Similarly, Figure 7b illustrates an on-track prediction for a child in the urban area. Positive contributing factors include the child’s exposure to three non-violent disciplinary activities, the child’s age (48 months), the mother’s age, the quality of the physical home environment, the highest wealth index quintile, residence in the Southwest zone, maternal secondary education, and engagement in three cognitive caregiving activities. In this case, physical violence remains the only factor that exerts a negative influence, slightly lowering the prediction. DISCUSSION This study applied advanced ensemble machine learning (ML) approaches to identify and rank 15 candidate predictors associated with early childhood development (ECD). The primary objective was to classify children as either on-track or off-track in their developmental progress. The interpretability of the model was enhanced through the application of SHapley Additive exPlanations (SHAP), a method rooted in feature attribution, which provided insights into the contributions of individual predictors towards the model’s decision-making process. Fifteen predictors were analysed and ranked based on their relative importance in determining whether a child’s development was on-track or off-track. KernelSHAP was employed to provide both global and local interpretations of model predictions, revealing key interactions between features. In relation to Bronfenbrenner’s PPCT Bioecological Theory of Human Development, child age, a key person factor, consistently ranked as a top predictor of early childhood development (ECD) across all contexts. The high ranking of age within a brief three-year period (2–5 years) highlights the critical role of early growth in shaping development. This supports the Matthew effect, where early disparities tend to widen over time (Bast & Reitsma, 1998 ; Stanovich, 1986 ). Interestingly, as children age, their likelihood of being on track decreases due to interactions with other predictors. In rural and urban areas, age interacted with factors such as child functioning difficulties, parental wealth, geopolitical zone, cognitive and socioemotional caregiving, physical violence, crowding (rural), and maternal education (urban). Within the PPCT bioecological framework, two key microsystem processes were found to be significant determinants of early childhood development (ECD). Cognitive caregiving emerged as a critical predictor, underscoring the role of active and supportive parenting in fostering developmental progress. Prior research (Bizzego et al., 2022 ) has established a strong association between cognitive caregiving and enhanced cognitive and verbal abilities. In both rural and urban contexts, cognitive caregiving ranked fifth, with increased exposure positively correlating with a child's likelihood of being developmentally on track. This study further identified important interactions within the ecological system. Cognitive caregiving demonstrated an interaction with child age, indicating that as children grow, the benefits of cognitive stimulation become more pronounced. Additionally, an interaction with caregiver education was observed, suggesting that maternal educational attainment significantly influences the extent of cognitive stimulation provided. Caregiver education, ranked fourth overall, plays a crucial role in cognitive enrichment activities such as book reading and storytelling (DeTemple & Snow, 2003 ). These findings align with those of Bornstein and Leventhal ( 2015 ), who reported that children in households with higher parental education levels are more likely to receive cognitive stimulation, thereby promoting positive developmental outcomes. While caregiver education was ranked sixth in rural areas—likely due to lower educational attainment—it was ranked second in urban settings, reflecting its greater influence in more educationally advantaged environments. Despite these contextual variations, caregiver education remained a significant determinant of ECD across both rural and urban settings. Socioemotional caregiving ranked low across all contexts in this study. However, its interaction with cognitive caregiving in urban settings revealed a significant influence on early childhood development (ECD). Higher levels of socioemotional support—including affectionate engagement and emotional responsiveness—amplify the benefits of cognitive caregiving (e.g., reading, counting, problem-solving) (Britto et al., 2017 ; Black et al., 2017 ). As cognitive caregiving increases, SHAP values for socioemotional caregiving also rise, indicating that emotional support enhances the impact of intellectual stimulation (Shonkoff & Phillips, 2000 ). Conversely, lower socioemotional caregiving, even with moderate cognitive stimulation, yields diminished developmental outcomes. This aligns with research emphasizing that cognitive skills alone are insufficient without emotional security and social bonding (Heckman, 2006 ; Walker et al., 2011 ). Emotionally secure children demonstrate greater problem-solving abilities, resilience, and adaptability (Ginsburg, 2007 ) This interaction supports Bronfenbrenner’s ecological systems theory, which highlights the interconnected influence of caregiving environments on child development (Bronfenbrenner & Morris, 2006 ). A balanced approach—combining socioemotional nurturing with cognitive stimulation—fosters emotional intelligence, self-regulation, and learning readiness (Raver et al., 2007 ). Moreover, consistent emotional support has been linked to intrinsic motivation for learning, long-term academic success, and social competence (Jones & Bouffard, 2012 ). The quality of the physical home environment emerged as the third most significant predictor of early childhood development (ECD) in Nigeria. This finding aligns with existing research, emphasizing the role of household resources in fostering developmental progress. Limited financial means often result in homes with fewer amenities and structural deficiencies, adversely affecting child health and well-being (World Health Organization, 2021 ). Children from economically disadvantaged backgrounds experience environments with reduced cognitive stimulation and supportive interactions, negatively impacting their cognitive and socioemotional development (Bronfenbrenner & Morris, 2006 ). Thus, the physical home environment serves as a critical proximal microsystem influencing child development. From the PPCT framework perspective, interactions between children and their environments are shaped by sociodemographic factors, including household income, parental education, and caregiver mental health. This study identified an interaction between home environment quality and wealth index in rural areas, where wealthier households provide superior living conditions that enhance ECD. In contrast, in urban areas, larger household sizes were associated with lower SHAP values, indicating that overcrowding may hinder developmental outcomes. These findings corroborate literature emphasizing the home environment's pivotal role in childhood development (Alkire & Foster, 2011 ; Bradley, 2020 ). Research highlights that material deprivation undermines developmental progress (Aguilar & Sumner, 2019 ), while nurturing environments promote cognitive and socioemotional growth (Bradley, 2020 ). This study provides empirical support for anti-poverty initiatives aimed at enhancing ECD by equipping families with the financial resources to create healthier, more supportive home environments for young children. Child functioning difficulties emerged as the seventh most influential predictor of early childhood development (ECD) in this study. Developmental progress relies on acquiring sensory, motor, cognitive, communication, and socio-emotional skills (WHO, 2012). Higher levels of functioning difficulties were associated with negative SHAP values, indicating an increased risk of developmental delays. This aligns with the logistic regression results, reinforcing the role of environmental interactions in shaping child development (Lynch & Hanson, 2004). A key factor in this interaction is caregiver-child relationships, particularly the influence of disciplinary practices, violence, and abuse. Children with disabilities are at a three to four times higher risk of experiencing violence (Jones et al., 2012), with parents more likely to use physical punishment (WHO, 2012). The SHAP dependence plot highlights the impact of psychological aggression, where verbal threats and punitive measures exacerbate developmental risks. Conversely, non-violent caregiving practices, such as explaining misbehaviour or revoking privileges, correlate with higher SHAP values, suggesting a protective effect on ECD. Findings support attachment theory, emphasizing that secure, nurturing environments foster emotional stability and cognitive resilience. Positive caregiving enhances neural plasticity, promoting adaptation and developmental progress (Bornstein et al., 2022 ). Interventions should focus on reducing psychological aggression and promoting non-violent caregiving strategies to mitigate disparities in developmental outcomes and support children with functional difficulties. CONCLUSION This study highlights the complex factors influencing early childhood development (ECD) in rural and urban settings, using ensemble machine learning models and SHapley Additive exPlanations (SHAP) to identify and rank 15 key predictors. The findings align with Bronfenbrenner’s PPCT Bioecological Theory, emphasizing the dynamic interactions between individual and environmental factors. Child age emerged as the most influential predictor, with its impact shaped by other variables such as child functioning difficulties, caregiver education, and cognitive caregiving. Cognitive caregiving and caregiver education were significant across contexts, with maternal education amplifying cognitive stimulation efforts—ranking higher in urban areas due to greater educational access. Rural-urban disparities were evident, with rural wealth enhancing home environments and urban overcrowding negatively affecting child development. Socioemotional caregiving, while ranking lower overall, positively interacted with cognitive caregiving in urban areas, reinforcing the link between emotional support and cognitive growth. Child functioning difficulties also influenced ECD, with developmental delays exacerbated by psychological aggression. This highlights the need for non-violent caregiving strategies to foster resilience and cognitive adaptability. Ultimately, the study calls for targeted interventions that address rural-urban gaps by improving caregiver education, promoting holistic caregiving practices, enhancing home environments, and supporting children with functional challenges. A nuanced, context-aware approach is essential to ensure equitable developmental opportunities for all children. This study possesses several notable strengths. Firstly, the application of ensemble machine learning models enhances predictive accuracy by capturing complex, non-linear relationships between multiple predictors. This approach surpasses the limitations of conventional statistical methods, offering a more nuanced understanding of early childhood development (ECD) determinants (Shukla et al., 2022 ). Additionally, the use of SHapley Additive exPlanations (SHAP) significantly strengthens the study by providing both global and local interpretability of model outputs. This fosters transparency and aids in understanding predictor importance, bridging the gap between complex algorithms and practical application. Moreover, the study's theoretical grounding in Bronfenbrenner’s PPCT Bioecological Theory of Human Development adds robustness to the findings. This alignment with established human development theories ensures that the study's conclusions are both conceptually sound and empirically supported. Another key strength lies in the study’s context-specific insights, which highlight rural-urban disparities in ECD predictors. By examining these environmental variations, the study offers tailored insights that support context-sensitive policymaking and intervention strategies (LeCroy et al., 2021 ). Despite these strengths, several limitations warrant consideration. Firstly, data limitations may have influenced the predictive accuracy of the models. While machine learning offers advanced analytic capabilities, its reliance on the quality and scope of input data means that unmeasured confounders—such as parental mental health or community-level resources—could affect developmental outcomes. Secondly, the study’s cross-sectional design restricts the ability to establish causality. Although SHAP values highlight predictor importance, they do not imply direct cause-and-effect relationships. Longitudinal data would better capture developmental trajectories over time, offering deeper insights into how early childhood factors interact dynamically. Building on these strengths and limitations, several avenues for future research emerge. Firstly, adopting longitudinal designs would enable researchers to track developmental trajectories over time, offering a more dynamic understanding of how predictors influence ECD (Patel et al., 2021 ). Longitudinal studies can illuminate the temporal relationships between variables, enhancing causal inferences. Moreover, integrating multimodal data—such as neuroimaging and genetic markers—alongside traditional sociodemographic predictors could further enrich model accuracy. This holistic approach may uncover hidden determinants of ECD, contributing to more comprehensive models (Shukla et al., 2022 ). Declarations ETHICS APPROVAL AND CONSENT TO PARTICIPATE This study utilized secondary data obtained from the Multiple Indicator Cluster Survey (MICS). Access to the dataset was granted upon request, following the submission of a formal email to the MICS team, clearly stating the objectives and intended use of the data. Ethical approval was not required, as the data is publicly available and anonymized, with no identifiable information about the participants. CONSENT FOR PUBLICATION Not applicable AVAILABILITY OF DATA AND MATERIALS The datasets analyzed during the current study are available in the UNICEF MICS repository, accessible via https://mics.unicef.org/ upon reasonable request and registration. COMPETING INTERESTS None FUNDING This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. AUTHORS' CONTRIBUTIONS ODG conceptualized the study, conducted the data processing and analysis, interpreted the findings, and drafted the manuscript. JOA provided academic supervision, contributed to refining the study design and methodology, offered critical revisions to the manuscript, and approved the final version. BMG contributed to the interpretation of findings, and provided substantial feedback on manuscript revision. ACKNOWLEDGEMENTS Special thanks the UNICEF MICS team for providing access to the dataset. Special thanks to Prof A. Adebowale, Dr. R.F. Afolabi and other faculty members for their academic guidance throughout the research. References Aguilar, G. R.,& Sumner,A. (2019).Who are the world’s poor? A new profile of global multidimensional poverty (Working Paper 499). Center for Global Development . www.cgdev.org/sites/default/files/who-are-worlds-poor-new-profile-global-multidimensionalpoverty.pdf Alkire, S., and Foster, J. 2011. Counting and multidimensional poverty measurement. The Journal of Public Economics, 95, 476–487. https://doi.org/10.1016/j.jpubeco.2010.11.006 Azur, M. J., Stuart, E. A., Frangakis, C., and Leaf, P. J. 2011. Multiple imputation by chained equations: what is it and how does it work? International Journal of Methods in Psychiatric Research , 20 (1), 40–49. https://doi.org/10.1002/mpr.329 Bast, J., and Reitsma, P. 1998.Analyzing the development of individual differences in terms of Matthew effects in reading: Results from a Dutch longitudinal study. Developmental Psychology , 34 (6), 1373–1399. https://doi.org/10.1037/0012-1649.34.6.1373 Bizzego, A., Gabrieli, G., Lim, M., Rothenberg, W. A., Bornstein, M. H., and Esposito, G. 2022. Predictors of early childhood development: A machine learning approach. In Parenting and Child Development in Low- and Middle-Income Countries (pp. 210–239). https://doi.org/10.4324/9781003044925-6 Black, M. M., Walker, S. P., Fernald, L. C. H., Andersen, C. T., DiGirolamo, A. M., Lu, C., McCoy, D. C., Fink, G., Shawar, Y. R., Shiffman, J., Devercelli, A. E., Wodon, Q. T., Vargas-Barón, E., and Grantham-McGregor, S. 2016. Early childhood development coming of age: science through the life course. The Lancet , 389 (10064), 77–90. https://doi.org/10.1016/s0140-6736(16)31389-7 Black, M. M., Walker, S. P., Fernald, L. C., Andersen, C. T., DiGirolamo, A. M., Lu, C., McCoy, D. C., Fink, G., Shawar, Y. R., Shiffman, J., Devercelli, A. E., Wodon, Q. T., Vargas-Barón, E., Grantham-McGregor, S., and Lancet Early Childhood Development Series Steering Committee. 2017. Early childhood development coming of age: Science through the life course. The Lancet , 389 (10064), 77–90. https://doi.org/10.1016/S0140-6736(16)31389-7 Bornstein, M. H., and Leventhal,T. (Eds.). 2015. Ecological settings and processes in developmental systems.Volume 4 of the handbook of child psychology and developmental science (Editor-in- Chief: R. M. Lerner, 7th ed.).Wiley. Bornstein, M. H., Bradley, R. H., Rothenberg, W. A., Deater-Deckard, K., Esposito, G., Lansford, J. E., Zietz, S., Putnick, D. L., and Bizzego, A. 2022. Parenting and child development in Low- and Middle-Income countries. In Routledge eBooks . https://doi.org/10.4324/9781003044925 Bradley, R. H. 2020. The child’s environment. In Part of a book series: M. H. Bornstein (Ed.), Elements in child development . Cambridge University Press. https://doi.org/10.1017/9781108866040 Bradley, R. H. 2020. The child’s environment. In Part of a book series: M. H. Bornstein (Ed.), Elements in child development . Cambridge University Press. https://doi.org/10.1017/9781108866040 Britto, P. R., Lye, S. J., Proulx, K., Yousafzai, A. K., Matthews, S. G., Vaivada, T., Perez-Escamilla, R., Rao, N., Ip, P., Fernald, L. C. H., MacMillan, H., Hanson, M., Wachs, T. D., Yao, H., Yoshikawa, H., Cerezo, A., Leckman, J. F., Bhutta, Z. A., & Early Childhood Development Interventions Review Group, for the Lancet Early Childhood Development Series Steering Committee (2017). Nurturing care: promoting early childhood development. Lancet (London, England) , 389 (10064), 91–102. https://doi.org/10.1016/S0140-6736(16)31390-3 Bronfenbrenner, U., and Morris, P. A. 2006.The bioecological model of human development. In R. M. Lerner (Ed.),Theoretical models of human development.Volume 1 of the handbook of child psychology (Editors-in-Chief:W. Damon and R. M. Lerner, 6th ed., pp. 793–828).Wiley. https://doi.org/10.1002/9780470147658.chpsy0114 De Temple, J., and Snow, C. E. 2003. Learning words from books. In A. van Kleeck, S. A. Stahl, and E. B. Bauer (Eds.), On reading books to children (pp. 16–36). Erlbaum. Erickson, Nick, Mueller J, Shirkov A., Zhang H., Larroy P ., Li M., and Smola A. 2020 "AutoGluon-Tabular: Robust and Accurate AutoML for Structured Data." arXiv preprint arXiv:2003.06505 Galasso, E., and Wagstaff, A. 2018. The aggregate income losses from childhood stunting and the returns to a nutrition intervention aimed at reducing stunting. Economics and Human Biology , 34 , 225–238. https://doi.org/10.1016/j.ehb.2019.01.010 Ginsburg, K. R. (2007). The importance of play in promoting healthy child development and maintaining strong Parent-Child bonds. PEDIATRICS , 119 (1), 182–191. https://doi.org/10.1542/peds.2006-2697 Heckman, J. J. 2006. Skill formation and the economics of investing in disadvantaged children. Science , 312 (5782), 1900-1902. Jones, S. M., and Bouffard, S. M. 2012. Social and emotional learning in schools: From programs to strategies. Social Policy Report , 26 (4), 1-33. Jones, S. M., and Bouffard, S. M. 2012. Social and emotional learning in schools: From programs to strategies. Social Policy Report , 26 (4), 1-33. Lange, G., Wodon, Q., and Carey, K. 2017. The Changing Wealth of Nations 2018: Building a Sustainable Future. In Washington, DC: World Bank eBooks . https://doi.org/10.1596/978-1-4648-1046-6 LeCroy, M., Kim, R. S., Stevens, J., Hanna, D., & Isasi, C. (2021). Identifying key determinants of childhood obesity: A narrative review of machine learning studies. Childhood Obesity. https://doi.org/10.1089/chi.2020.0324 Patel, D., Hall, G., Broadhurst, D. I., Smith, A., Schultz, A., & Foong, R. (2021). Does machine learning have a role in the prediction of asthma in children? Paediatric Respiratory Reviews. https://doi.org/10.1016/j.prrv.2021.06.002 Raver, C. C., Smith-Donald, R., Hayes, T., and Jones, S. M. 2007. Self-regulation across different contexts: Implications for learning and early academic success. Early Education and Development , 18 (2), 303-320. Richter, L. M., Daelmans, B., Lombardi, J., Heymann, J., Boo, F. L., Behrman, J. R., Lu, C., Lucas, J. E., Perez-Escamilla, R., Dua, T., Bhutta, Z. A., Stenberg, K., Gertler, P., Darmstadt, G. L., and Paper 3 Working Group and the Lancet Early Childhood Development Series Steering Committee 2017. Investing in the foundation of sustainable development: pathways to scale up for early childhood development. Lancet (London, England) , 389 (10064), 103–118. https://doi.org/10.1016/S0140-6736(16)31698-1 Shonkoff, J. P., and Phillips, D. A. 2000. "From neurons to neighborhoods: The science of early childhood development." National Academy Press. Shukla, P. K., Jain, S., & Kalra, S. (2022). Machine learning in childhood asthma prediction: A systematic review. 2022 10th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO) , 1–6. https://doi.org/10.1109/ICRITO56286.2022.9964849 Spencer, N., Raman, S., O'Hare, B., and Tamburlini, G. 2019. Addressing inequities in child health and development: towards social justice. BMJ paediatrics open , 3 (1), e000503. https://doi.org/10.1136/bmjpo-2019-000503 Stanovich, K. E. 1986. Matthew effects in reading: Some consequences of individual differences in the acquisition of literacy. Reading Research Quarterly , 21 (4), 360–407. https://doi.org/10.1598/RRQ.21.4.1 United Nations Children’s Fund 2023. The Early Childhood Development Index 2030: A new measure of early childhood development, UNICEF, New York. Walker, S. P., Wachs, T. D., Grantham-McGregor, S., et al. 2011. Inequality in early childhood: Risk and protective factors for early child development. The Lancet , 378 (9799), 1325-1338. World Health Organization 2012. Early childhood development and disability: A discussion paper. World Health Organization. 2021. World health statistics. Monitoring health for the SDGs . World Health Organization. Additional Declarations No competing interests reported. Supplementary Files SUPPLEMENTARYFILE.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 19 Mar, 2026 Reviewers agreed at journal 13 Mar, 2026 Reviewers agreed at journal 12 Mar, 2026 Reviews received at journal 11 Mar, 2026 Reviewers agreed at journal 06 Mar, 2026 Reviewers invited by journal 28 Jan, 2026 Editor invited by journal 21 Jan, 2026 Editor assigned by journal 20 Jan, 2026 Submission checks completed at journal 20 Jan, 2026 First submitted to journal 15 Jan, 2026 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-8615011","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":582546450,"identity":"3bcc8be9-a1fd-413a-9153-b51eec4c6913","order_by":0,"name":"Obaloluwa David GBADEGESIN","email":"data:image/png;base64,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","orcid":"","institution":"University of Ibadan","correspondingAuthor":true,"prefix":"","firstName":"Obaloluwa","middleName":"David","lastName":"GBADEGESIN","suffix":""},{"id":582546451,"identity":"7fb35072-8ed8-4c41-9cdf-96923752892d","order_by":1,"name":"Babatunde Makinde Gbadebo","email":"","orcid":"","institution":"University of Ibadan","correspondingAuthor":false,"prefix":"","firstName":"Babatunde","middleName":"Makinde","lastName":"Gbadebo","suffix":""},{"id":582546452,"identity":"b2cc3f86-0d00-4f3c-a2b8-c8f65903b205","order_by":2,"name":"Joshua Odunayo AKINYEMI","email":"","orcid":"","institution":"University of Ibadan","correspondingAuthor":false,"prefix":"","firstName":"Joshua","middleName":"Odunayo","lastName":"AKINYEMI","suffix":""}],"badges":[],"createdAt":"2026-01-16 03:53:09","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8615011/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8615011/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":101631040,"identity":"71900774-bd50-449c-852e-7599d4eb7503","added_by":"auto","created_at":"2026-02-02 05:28:17","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":300996,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eModel architecture of MSWEM\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8615011/v1/eb4d968efcb9abfcb9e73c2b.png"},{"id":101753680,"identity":"b9d416f8-2025-41ab-8581-e4a6a89768ab","added_by":"auto","created_at":"2026-02-03 10:40:32","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":104637,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eReceiver operating characteristics curve of MSWEM for validation set (a) and test set (b)\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8615011/v1/9f6b36ec4bb2fb7f98b0c8c6.png"},{"id":101631036,"identity":"c7a0b0b8-6192-4ce9-9f50-56bb5f4436f6","added_by":"auto","created_at":"2026-02-02 05:28:17","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":175341,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP Summary plot of the Multilayer Stacked Weighted Ensemble Model for Rural Area(A) and Urban Area (B)\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8615011/v1/f9e6b82072831e30b2908441.png"},{"id":101753681,"identity":"4d896dc7-0eaa-4b12-9e4e-f5908e5d27b3","added_by":"auto","created_at":"2026-02-03 10:40:32","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":344202,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP dependence plot showing interaction between variables in rural area.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8615011/v1/253c977d3d68e10a56b493a4.png"},{"id":101631043,"identity":"26d4f451-1b02-4f1b-9c60-eebd66a802a3","added_by":"auto","created_at":"2026-02-02 05:28:18","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":194824,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP dependence plot showing interaction between variables in urban area.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8615011/v1/4a98926d21ac64df15bbf004.png"},{"id":101753440,"identity":"fe9d6ddb-f278-4969-b074-718a50a88188","added_by":"auto","created_at":"2026-02-03 10:40:03","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":122373,"visible":true,"origin":"","legend":"\u003cp\u003ea: SHAP force plot explaining an off-track prediction for a child from a rural area.\u003c/p\u003e\n\u003cp\u003eFigure 6b: SHAP force plot explaining an on-track prediction for a child from a rural area.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8615011/v1/4e93f0262e8509c66de18d11.png"},{"id":101943175,"identity":"57d98be8-f170-4802-aa3c-c1af4897df0f","added_by":"auto","created_at":"2026-02-05 09:40:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2210304,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8615011/v1/2d12c68c-250f-480f-8fe9-e82f1115abbf.pdf"},{"id":101752948,"identity":"24a9d97b-5c63-4fb8-9b1c-d086f0a388b2","added_by":"auto","created_at":"2026-02-03 10:38:19","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":16653,"visible":true,"origin":"","legend":"","description":"","filename":"SUPPLEMENTARYFILE.docx","url":"https://assets-eu.researchsquare.com/files/rs-8615011/v1/dc29b5d71c96f04cf775bbd2.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Rural-Urban Disparities in Early Childhood Development: Insights from Interpretable Ensemble Machine Learning Model","fulltext":[{"header":"BACKGROUND","content":"\u003cp\u003eEarly childhood development (ECD) is a cornerstone of human potential, laying the groundwork for lifelong health, education, and economic success. It primarily focuses on children aged eight years and younger, encompassing the development of motor, language, cognitive, socio-emotional, and self-regulatory skills (UNICEF, 2023). These abilities develop during the early years of life, a unique and critical period characterized by extraordinary brain plasticity\u0026mdash;the capacity of the brain to adapt and grow in response to experiences. This heightened plasticity offers both immense opportunities for growth and profound vulnerabilities, as the brain\u0026rsquo;s development is heavily influenced by early experiences (Spencer et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eResearch demonstrates that timely support during this critical period yields far greater benefits than efforts made later in life to compensate for early disadvantages (Britto et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). For this reason, ECD interventions are vital for fostering human development and breaking intergenerational cycles of poverty. Despite significant strides in child survival, such as a 53% reduction in under-five mortality between 1990 and 2015, the burden of developmental risk remains high.\u003c/p\u003e \u003cp\u003eApproximately 250\u0026nbsp;million children under five years of age in low- and middle-income countries (LMICs) are at risk of not reaching their full developmental potential, a figure that rises to two-thirds in Sub-Saharan Africa (Black et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). These risks are compounded by factors such as extreme poverty and stunting, which not only hinder individual potential but also impose significant societal costs. For instance, the mean annual loss in adult income due to these risks is estimated at 26%, perpetuating poverty and stunting economic growth (Richter et al., 2017). Eliminating stunting alone could have increased per capita GDP by 7%, according to World Bank estimates (Galasso \u0026amp; Wagstaff, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe risks extend beyond poverty and malnutrition to include factors like inadequate caregiving practices, parental discipline and violence, the physical home environment, and child functioning difficulties. These cumulative risks threaten not just individual developmental outcomes but also the human capital that drives societal progress. As such, ECD forms the foundation of human capital, accounting for a significant portion of the wealth disparities between nations (Lange et al., 2018).\u003c/p\u003e \u003cp\u003eExisting research on ECD in Nigeria often focuses on factors such as poverty and malnutrition while context-specific determinants like caregiving practices, parental discipline, exposure to violence, and the physical home environment are scarcely investigated. This limited scope fails to capture the complex interplay of factors influencing child development and constrains efforts to design effective, evidence-based interventions tailored to the Nigerian context. Furthermore, public health and social systems in Nigeria are often unable to identify children at risk of developmental delays early for prompt and effective interventions.\u003c/p\u003e \u003cp\u003eTraditional approaches to analysis of ECD risk factors have relied on linear models that emphasize individual predictors but fail to account for non-linear relationships and interactions among variables. While machine learning (ML) offers advanced predictive capabilities and the potential to uncover hidden patterns in large datasets, its application in ECD research in Nigeria remains minimal. The \"black box\" nature of many ML models further complicates their adoption in public health, where interpretability is essential for translating findings into actionable policies. Emerging approaches such as explainable AI (XAI) address this challenge, offering tools to enhance the interpretability of complex ML models.\u003c/p\u003e \u003cp\u003eThis study aims to bridge these gaps by leveraging interpretable ensemble machine learning models for early childhood development and analyse the socio-demographic, environmental, and caregiving factors influencing ECD outcomes. The strong alignment of this study\u0026rsquo;s findings with Bronfenbrenner\u0026rsquo;s PPCT Bioecological Theory of Human Development is also evident. By incorporating underexplored determinants, such as parental caregiving practices, home environments, and exposure to violence, this research will provide a more comprehensive understanding of ECD disparities in Nigeria. The findings will inform the development of targeted interventions and policies to address these disparities, reduce developmental inequities, and enhance developmental outcomes for Nigerian children.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eDATA SOURCE AND DESCRIPTION\u003c/h2\u003e \u003cp\u003eThis study utilizes data from the sixth round of the Nigeria Multiple Indicator Cluster Survey conducted in 2021. The study population consists of children aged 2 to 5 years and their families, drawn from the nationally representative sample of the 2021 MICS 6 dataset for Nigeria. The data was collected through a multi-stage stratified cluster sampling approach. This age cohort was selected to align with the study\u0026rsquo;s objective of examining early childhood development and predictors.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSTUDY VARIABLES\u003c/h3\u003e\n\u003cp\u003eThe new measure of early childhood development which is the Early Childhood Development Index 2030 was estimated from a list of 20 questions asked in the 2021 MICS6 round according to age groups and this was used to generate a single summative score (United Nations Children\u0026rsquo;s Fund, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). ECDI2030 is given as the proportion of children aged 24 to 59 months who are developmentally on track. The dependent variable was thus obtained by generating a variable that identifies children with on-track development according to the pre-defined age-specific cut-scores based on total milestones achieved from the 20 items (UNICEF, 2023).This variable generated was used to report the proportion of children aged 24 to 59 months who were developmentally on-track in health, learning and psychosocial wellbeing. Independent variables included age of child in month, gender, age of mother/caregiver, education, wealth index, zone, household characteristics, caregiving practices, parental discipline and violence practices, child functioning difficulties. Independent variables were selected based on existing literature on early childhood development (Bornstein et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e\n\u003ch3\u003eDATA MANAGEMENT AND ANALYSIS\u003c/h3\u003e\n\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eData Preprocessing\u003c/h2\u003e \u003cp\u003eThe MICS data was extensively pre-processed to ensure that it is suitable for machine learning analysis. To address missing data in this study, a multiple imputation technique using chained equations was used, following the approach described by Azur et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). This method involves running several regression models to predict missing values based on other observed variables in the dataset. The type of regression model used depends on the nature of the variable being imputed. Likewise, class imbalance was observed in the target variable -whether a child is developmentally on-track or off-track, and this was handled using the Synthetic Minority Oversampling Technique (SMOTE). SMOTE is an effective oversampling method that generates synthetic data points for the minority class, helping to create a balanced dataset and improve model performance.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eMODEL DEVELOPMENT\u003c/h3\u003e\n\u003cp\u003eThis study utilized advanced ensemble machine learning techniques, including stacking and bagging, to develop predictive models with superior performance and reduced variance. Stack ensembling involves training multiple base models and combining their predictions using a meta-model (stacker). Multi-layer stacking extends this by creating hierarchical layers of stackers, where each layer builds upon the predictions of the previous one. In this approach, we have base models and stacker models and AutoGluon\u0026rsquo;s framework was used to introduce novel enhancements such as reusing base model architectures, incorporation of original features, ensemble selection in the final layer using a greedy weighted algorithm as highlighted by Erickson et al., (2020).\u003c/p\u003e \u003cp\u003eThe dataset was partitioned into three subsets comprising 80% for training, 10% for validation, and 10% for testing. To optimize the use of training data and reduce prediction variance, \u003cem\u003ek\u003c/em\u003e-fold bagging was employed across all layers of the ensemble architecture, in line with the approach described by Erickson et al. (2020). This method involves dividing the training data into \u003cem\u003ek\u003c/em\u003e equal folds, sequentially training the model on \u003cem\u003ek\u0026ndash;1\u003c/em\u003e folds and validating on the remaining fold. Through this process, every data point is eventually utilized for both training and validation, thereby improving the model's generalization capacity and minimizing overfitting. To further stabilize predictions, repeated \u003cem\u003ek\u003c/em\u003e-fold bagging was implemented. Overall, the total number of models trained within the ensemble is given by the expression \u003cb\u003eM \u0026times; N \u0026times; K\u0026thinsp;+\u0026thinsp;1\u003c/b\u003e, where \u003cem\u003eM\u003c/em\u003e denotes the number of stacking layers (excluding the final aggregation layer), \u003cem\u003eN\u003c/em\u003e is the number of model types per layer, \u003cem\u003eK\u003c/em\u003e is the number of bagging folds, and the additional \u003cem\u003e1\u003c/em\u003e corresponds to the final meta-model trained using a Greedy Weighted Averaging algorithm. The resulting ensemble model was subsequently used to generate predictions for both rural and urban populations.\u003c/p\u003e \u003cp\u003eAccording to Lundberg and Lee (2018), the best explanation for a simple model is the model itself, as it inherently represents its decision process in a straightforward manner. However, for complex models, which lack inherent interpretability, an explanation model must be used. This explanation model \u003cem\u003eg\u003c/em\u003e is defined as an interpretable approximation of the original prediction model \u003cem\u003ef\u003c/em\u003e, focusing on local explanations based on a specific input \u003cem\u003ex\u003c/em\u003e. Kernel SHAP was employed in this study to achieve model interpretability. The performance of the classification models was assessed using standard evaluation metrics, including Accuracy, Precision, Recall (Sensitivity), F1 Score, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and the Area Under the Receiver Operating Characteristic Curve (ROC_AUC).\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003e\u003cstrong\u003eSociodemographic characteristics of participants\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData from 15,899 households/children were analysed by filtering the MICS dataset to include children aged 24 to 59 months. Of these, the average age of the children included in the study was 41.3 months (SD = 10.3), with a relatively even distribution by gender: 49.5% male and 50.4% female. The sociodemographic characteristics of primary caregivers examined in this study were age and educational status as shown in Table 1. On average, primary caregivers were 31.2 years old (SD = 7.1). Also, Table 1 also revealed that the educational backgrounds of primary caregivers varied with 42.3% reported having no formal education, 15.9% completed only primary education, 6.7% attained junior secondary school education while 25.1% completed senior secondary education, and 10.0% received higher/tertiary education.\u003c/p\u003e\n\u003cp\u003eIn terms of household characteristics, the average household size in the sample was 7.9 individuals (SD = 4.0). On average, households had 4.95 children under the age of 18 years (SD = 2.98), with 2.11 children under the age of five years (SD = 1.1). Additionally, a newly created variable estimating household crowding indicated that, on average, 3.39 people shared a bedroom (SD = 1.47).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOn average, families owned 3.57 (SD = 1.38) out of five possible electronic assets, 1.88 (SD = 1.11) out of seven possible family amenities, and 1.92 (SD = 1.04) out of five possible learning materials. Regarding housing quality, 31.5% of families used unimproved water sources, and 24.5% lacked adequate or used unimproved toileting facilities. Additionally, 34.7% of homes had natural floors, 13.4% had natural roofs, and 35.4% had natural exterior walls. A significant majority (96.1%) of families cooked with solid, less healthy fuel sources, and 45.5% cooked outside the home.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e: \u003cstrong\u003eBackground Characteristics of All Participants\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean (SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eFrequency (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eChild characteristics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN = 15899\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean age of child\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e41.3 (10.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eGender\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e49.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e50.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrimary Caregivers\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean Age\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e31.2 (7.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eEducation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNo formal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e42.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrimary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eJunior Secondary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSenior Secondary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e25.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigher / Tertiary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eHousehold characteristics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eHousehold size\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7.90 (4.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNumber of children under 18 years in an household\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e4.95 (2.98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNumber of children under 5 years in an household\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2.11 (1.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3.39 (1.47)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eEarly Childhood Development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe new measure of early childhood development which is the Early Childhood Development Index 2030 was estimated from a list of 20 question asked in the 2021 MICS6 round according to age groups. The total milestones according to specific age groups was used to classify under five children as either developmentally on-track or off-track according to the ECDI2030 Technical Manual (UNICEF, 2023). The result revealed that in a sample of 15899 children in Nigeria, the total on-track children is 38.2% . The result was further disaggregated by sex, area, mother\u0026rsquo;s education, functional difficulties, wealth index and geopolitical zone as presented in Table 2.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2: Percentage of Children aged 2 \u0026ndash; 4 years who are developmentally on track in health, learning and psychosocial well-being in Nigeria\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 35px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePercentage of Children who are developmentally on-track\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNumber of children aged 2 \u0026ndash; 4 years\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 35px;\"\u003e\n \u003cp\u003e38.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e15899\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSex\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eFemale\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eMale\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 35px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e39.1%\u003c/p\u003e\n \u003cp\u003e37.3%\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e7904\u003c/p\u003e\n \u003cp\u003e8039\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eArea\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eRural\u003c/p\u003e\n \u003cp\u003eUrban\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 35px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e29.2%\u003c/p\u003e\n \u003cp\u003e55.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e10552\u003c/p\u003e\n \u003cp\u003e5391\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMother\u0026rsquo;s education\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eNone\u0026nbsp;\u003c/p\u003e\n \u003cp\u003ePrimary\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eJunior Secondary\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eSenior Secondary\u003c/p\u003e\n \u003cp\u003eHigher / tertiary\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 35px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e22.7%\u003c/p\u003e\n \u003cp\u003e37.5%\u003c/p\u003e\n \u003cp\u003e41.5%\u003c/p\u003e\n \u003cp\u003e51.9%\u003c/p\u003e\n \u003cp\u003e69.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e6749\u003c/p\u003e\n \u003cp\u003e2531\u003c/p\u003e\n \u003cp\u003e1071\u003c/p\u003e\n \u003cp\u003e3995\u003c/p\u003e\n \u003cp\u003e1598\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFunctional disabilities\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eHas functional disabilities\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eHas no functional disabilities\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 35px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e14.3%\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e39.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e671\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e15272\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWealth index quintile\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003ePoorest\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eRichest\u003c/p\u003e\n \u003cp\u003eSecond\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eMiddle\u003c/p\u003e\n \u003cp\u003eFourth\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 35px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e22.2%\u003c/p\u003e\n \u003cp\u003e68.0%\u003c/p\u003e\n \u003cp\u003e27.8%\u003c/p\u003e\n \u003cp\u003e37.0%\u003c/p\u003e\n \u003cp\u003e52.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 34px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e4138\u003c/p\u003e\n \u003cp\u003e2335\u003c/p\u003e\n \u003cp\u003e3698\u003c/p\u003e\n \u003cp\u003e3098\u003c/p\u003e\n \u003cp\u003e2675\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 0px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGeopolitical zone\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eNorth West\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eNorth East\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eNorth Central\u003c/p\u003e\n \u003cp\u003eSouth South\u003c/p\u003e\n \u003cp\u003eSouth East\u003c/p\u003e\n \u003cp\u003eSouth West\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 0px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e27.0%\u003c/p\u003e\n \u003cp\u003e26.9%\u003c/p\u003e\n \u003cp\u003e34.2%\u003c/p\u003e\n \u003cp\u003e53.1%\u003c/p\u003e\n \u003cp\u003e53.2%\u003c/p\u003e\n \u003cp\u003e65.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 0px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e5651\u003c/p\u003e\n \u003cp\u003e2811\u003c/p\u003e\n \u003cp\u003e2197\u003c/p\u003e\n \u003cp\u003e1657\u003c/p\u003e\n \u003cp\u003e1545\u003c/p\u003e\n \u003cp\u003e2082\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\u003cp\u003e\u003cstrong\u003eEVALUATION OF THE MULTILAYER STACKED WEIGHTED ENSEMBLE MODEL ON THE TRAINING SET, VALIDATION SET AND TEST SET.\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 1 shows the model architecture of a stacked ensemble model, highlighting the steps taken to achieve optimal predictive accuracy. This model was developed using multiple layers of base learners, each contributing to the final ensemble prediction. The performance of the Multilayer Stacked Weighted Ensemble Model was evaluated using various metrics on the training, validation, and test datasets. The model achieved an accuracy of 100% on the training set, demonstrating its ability to correctly classify nearly all cases in the training data. On the validation set, accuracy became 83.7%, and on the test set, it increased to 85.2%. Also, the assessment of the model predicting power was compared using the validation and test set and this is shown in Figure 2. Performances were optimum in both validation and test set.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClassification Performance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTable 3 shows the predictive performance of the newly developed model across rural, urban areas in Nigeria. A total of 24.7% were predicted to be on-track in their early childhood development in rural areas while 46.4% of children were predicted to be on-track in urban areas in Nigeria.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3: Performance of MSWEM in classifying on-track and off-track in rural area and urban areas.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRural\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpecificity\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1-score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOff track (0)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e98%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e91%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e94%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOn track (1)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e77%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e93%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e84%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy = 92%\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eROC_AUC = 87%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUrban\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOff track (0)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e92%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e88%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e90%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOn track (1)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e87%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 30px;\"\u003e\n \u003cp\u003e91%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e89%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy = 90%\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eROC_AUC = \u0026nbsp;90%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eMODEL INTERPRETATION IN RURAL AND URBAN AREAS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 3 presents SHAP summary plots comparing feature importance in rural and urban areas.\u003c/p\u003e\n\u003cp\u003eIn both contexts, a child\u0026rsquo;s age is the most influential factor. In rural areas, maternal age plays a key role, with older mothers more likely to have children on track developmentally. In contrast, maternal education was the second most important factor in urban areas, where higher educational attainment strongly correlates with better child development. Although maternal education influences both contexts, it ranks lower in rural areas, reflecting lower educational attainment among rural mothers.\u003c/p\u003e\n\u003cp\u003eThe wealth index shows a stark contrast, ranking highly in urban areas but significantly lower in rural regions, indicating wealth concentration in cities. Interaction effects within urban areas further highlight the complex relationship between wealth and child development.\u003c/p\u003e\n\u003cp\u003eThe home environment is a critical predictor in both settings, though its variability is lower in rural areas. In urban areas, better home environments strongly correlate with improved child outcomes. Cognitive caregiving exhibits less variability in rural areas but remains a strong positive influence across both settings. Similarly, socioemotional caregiving ranks lower in both areas, yet SHAP plots reveal interactions underscoring its importance in shaping developmental outcomes.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eINTERACTION OF VARIABLES\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn rural areas, a negative relationship between age and SHAP values indicates that younger children (25\u0026ndash;35 months) as shown in figure 4 benefit more from age-related developmental gains, but this effect diminishes as they grow older. Functional difficulties consistently lower SHAP values, reducing development across all ages. Household wealth and geographic zone have a stronger influence on children under five, while older children from wealthier families and favourable zones (South-South and South-West) are more likely to be on track. Older mothers engage more in socioemotional activities, and higher cognitive care strongly enhances developmental outcomes, especially in older children. Socioemotional care has a greater impact on younger children but varies across age groups. Additional interactions exist between physical violence and child age, home environment and wealth index, and household crowding and child age.\u003c/p\u003e\n\u003cp\u003eConversely, SHAP dependence plots for children in urban areas shown in Figure 5 reveal that maternal education is a significant predictor of early childhood development (ECD) in urban areas. Interaction effects between maternal education and the child\u0026rsquo;s age were observed. The analysis indicates that higher SHAP values for maternal education were associated with children whose mothers possess higher levels of education. Higher SHAP values, which positively influence predictions of children being on track, correlate with increased levels of maternal education and the number of cognitive activities to which a child is exposed. Conversely, lower levels of exposure to cognitive caregiving were associated with decreased SHAP values, while higher exposure to socioemotional caregiving practices mitigated this effect. Furthermore, the analysis reveals that a smaller household size increases the SHAP values for the physical environment, thereby enhancing predictions of a child being on track. These findings underscore the critical role of maternal education, cognitive stimulation, and household composition in shaping developmental outcomes in urban contexts.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eINTERPRETATION OF THE MODEL IN RURAL AND URBAN AREAS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFollowing an examination of variable interactions, it became evident that further investigation into the contributions of each variable to predicting early childhood development (ECD) in rural and urban areas was necessary. An individualized explanation of how variables influence ECD predictions in these contexts is presented through the SHAP force plots in Figures 6 and 7. These plots provide detailed insights into model decision-making for rural and urban areas, respectively. In Figure 6, which focuses on the rural area, the SHAP force plot illustrates how various features contribute to the prediction for a single child. Features highlighted in red exert a positive influence on the final prediction, while those in blue negatively impact the outcome. The force plot represents a specific data point where the model predicts whether the child is on-track or off-track.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRURAL AREA\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe feature \u0026quot;Crowd\u0026quot; exhibits a SHAP value of 4.5, significantly elevating the prediction above the base value. Positive influences on the prediction include the child\u0026rsquo;s exposure to non-violent disciplinary practices, psychological aggression activities, the physical home environment, and the child\u0026apos;s age, all of which contribute to an increased likelihood of being on track. Conversely, factors that lowered the prediction include a large household size and the young age of the mother. In this instance, the prediction for the child was off-track, primarily driven by the negative impact of larger family sizes, residence in the Northeast zone (represented as 2), and the young maternal age.\u003c/p\u003e\n\u003cp\u003eFor children predicted to be on-track, the force plot reveals positive driving factors that elevated the prediction beyond the base value. These include exposure to two cognitive caregiving activities, residence in a wealthier household, engagement in two psychological aggression activities, the quality of the home environment, as well as the age of the mother and child. This individualized analysis underscores the intricate balance between positive and negative factors influencing ECD outcomes in rural contexts.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eURBAN AREA\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 7a presents an analysis of an off-track prediction for a child residing in an urban area. The SHAP force plot highlights the variables that contribute positively and negatively to the final prediction. Factors that push the prediction higher relative to the base value include the quality of the physical home environment, the presence of four household members, a middle wealth index quintile, residence in the Northeast zone, and a 23-year-old mother with secondary education who engages the child in two cognitive caregiving activities. Conversely, variables that lowered the prediction include the occurrence of physical violence, the child\u0026rsquo;s age (50 months), and exposure to only one socioemotional caregiving activity. These factors collectively diminish the likelihood of the child being on track.\u003c/p\u003e\n\u003cp\u003eSimilarly, Figure 7b illustrates an on-track prediction for a child in the urban area. Positive contributing factors include the child\u0026rsquo;s exposure to three non-violent disciplinary activities, the child\u0026rsquo;s age (48 months), the mother\u0026rsquo;s age, the quality of the physical home environment, the highest wealth index quintile, residence in the Southwest zone, maternal secondary education, and engagement in three cognitive caregiving activities. In this case, physical violence remains the only factor that exerts a negative influence, slightly lowering the prediction.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eThis study applied advanced ensemble machine learning (ML) approaches to identify and rank 15 candidate predictors associated with early childhood development (ECD). The primary objective was to classify children as either on-track or off-track in their developmental progress. The interpretability of the model was enhanced through the application of SHapley Additive exPlanations (SHAP), a method rooted in feature attribution, which provided insights into the contributions of individual predictors towards the model\u0026rsquo;s decision-making process.\u003c/p\u003e \u003cp\u003eFifteen predictors were analysed and ranked based on their relative importance in determining whether a child\u0026rsquo;s development was on-track or off-track. KernelSHAP was employed to provide both global and local interpretations of model predictions, revealing key interactions between features.\u003c/p\u003e \u003cp\u003eIn relation to Bronfenbrenner\u0026rsquo;s PPCT Bioecological Theory of Human Development, child age, a key person factor, consistently ranked as a top predictor of early childhood development (ECD) across all contexts. The high ranking of age within a brief three-year period (2\u0026ndash;5 years) highlights the critical role of early growth in shaping development. This supports the Matthew effect, where early disparities tend to widen over time (Bast \u0026amp; Reitsma, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; Stanovich, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e1986\u003c/span\u003e). Interestingly, as children age, their likelihood of being on track decreases due to interactions with other predictors. In rural and urban areas, age interacted with factors such as child functioning difficulties, parental wealth, geopolitical zone, cognitive and socioemotional caregiving, physical violence, crowding (rural), and maternal education (urban).\u003c/p\u003e \u003cp\u003eWithin the PPCT bioecological framework, two key microsystem processes were found to be significant determinants of early childhood development (ECD). Cognitive caregiving emerged as a critical predictor, underscoring the role of active and supportive parenting in fostering developmental progress. Prior research (Bizzego et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) has established a strong association between cognitive caregiving and enhanced cognitive and verbal abilities. In both rural and urban contexts, cognitive caregiving ranked fifth, with increased exposure positively correlating with a child's likelihood of being developmentally on track.\u003c/p\u003e \u003cp\u003eThis study further identified important interactions within the ecological system. Cognitive caregiving demonstrated an interaction with child age, indicating that as children grow, the benefits of cognitive stimulation become more pronounced. Additionally, an interaction with caregiver education was observed, suggesting that maternal educational attainment significantly influences the extent of cognitive stimulation provided. Caregiver education, ranked fourth overall, plays a crucial role in cognitive enrichment activities such as book reading and storytelling (DeTemple \u0026amp; Snow, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). These findings align with those of Bornstein and Leventhal (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2015\u003c/span\u003e), who reported that children in households with higher parental education levels are more likely to receive cognitive stimulation, thereby promoting positive developmental outcomes.\u003c/p\u003e \u003cp\u003eWhile caregiver education was ranked sixth in rural areas\u0026mdash;likely due to lower educational attainment\u0026mdash;it was ranked second in urban settings, reflecting its greater influence in more educationally advantaged environments. Despite these contextual variations, caregiver education remained a significant determinant of ECD across both rural and urban settings.\u003c/p\u003e \u003cp\u003eSocioemotional caregiving ranked low across all contexts in this study. However, its interaction with cognitive caregiving in urban settings revealed a significant influence on early childhood development (ECD). Higher levels of socioemotional support\u0026mdash;including affectionate engagement and emotional responsiveness\u0026mdash;amplify the benefits of cognitive caregiving (e.g., reading, counting, problem-solving) (Britto et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Black et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). As cognitive caregiving increases, SHAP values for socioemotional caregiving also rise, indicating that emotional support enhances the impact of intellectual stimulation (Shonkoff \u0026amp; Phillips, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2000\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eConversely, lower socioemotional caregiving, even with moderate cognitive stimulation, yields diminished developmental outcomes. This aligns with research emphasizing that cognitive skills alone are insufficient without emotional security and social bonding (Heckman, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Walker et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Emotionally secure children demonstrate greater problem-solving abilities, resilience, and adaptability (Ginsburg, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2007\u003c/span\u003e)\u003c/p\u003e \u003cp\u003eThis interaction supports Bronfenbrenner\u0026rsquo;s ecological systems theory, which highlights the interconnected influence of caregiving environments on child development (Bronfenbrenner \u0026amp; Morris, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). A balanced approach\u0026mdash;combining socioemotional nurturing with cognitive stimulation\u0026mdash;fosters emotional intelligence, self-regulation, and learning readiness (Raver et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). Moreover, consistent emotional support has been linked to intrinsic motivation for learning, long-term academic success, and social competence (Jones \u0026amp; Bouffard, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2012\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe quality of the physical home environment emerged as the third most significant predictor of early childhood development (ECD) in Nigeria. This finding aligns with existing research, emphasizing the role of household resources in fostering developmental progress. Limited financial means often result in homes with fewer amenities and structural deficiencies, adversely affecting child health and well-being (World Health Organization, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Children from economically disadvantaged backgrounds experience environments with reduced cognitive stimulation and supportive interactions, negatively impacting their cognitive and socioemotional development (Bronfenbrenner \u0026amp; Morris, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). Thus, the physical home environment serves as a critical proximal microsystem influencing child development.\u003c/p\u003e \u003cp\u003eFrom the PPCT framework perspective, interactions between children and their environments are shaped by sociodemographic factors, including household income, parental education, and caregiver mental health. This study identified an interaction between home environment quality and wealth index in rural areas, where wealthier households provide superior living conditions that enhance ECD. In contrast, in urban areas, larger household sizes were associated with lower SHAP values, indicating that overcrowding may hinder developmental outcomes.\u003c/p\u003e \u003cp\u003eThese findings corroborate literature emphasizing the home environment's pivotal role in childhood development (Alkire \u0026amp; Foster, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Bradley, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Research highlights that material deprivation undermines developmental progress (Aguilar \u0026amp; Sumner, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), while nurturing environments promote cognitive and socioemotional growth (Bradley, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). This study provides empirical support for anti-poverty initiatives aimed at enhancing ECD by equipping families with the financial resources to create healthier, more supportive home environments for young children.\u003c/p\u003e \u003cp\u003eChild functioning difficulties emerged as the seventh most influential predictor of early childhood development (ECD) in this study. Developmental progress relies on acquiring sensory, motor, cognitive, communication, and socio-emotional skills (WHO, 2012). Higher levels of functioning difficulties were associated with negative SHAP values, indicating an increased risk of developmental delays. This aligns with the logistic regression results, reinforcing the role of environmental interactions in shaping child development (Lynch \u0026amp; Hanson, 2004).\u003c/p\u003e \u003cp\u003eA key factor in this interaction is caregiver-child relationships, particularly the influence of disciplinary practices, violence, and abuse. Children with disabilities are at a three to four times higher risk of experiencing violence (Jones et al., 2012), with parents more likely to use physical punishment (WHO, 2012). The SHAP dependence plot highlights the impact of psychological aggression, where verbal threats and punitive measures exacerbate developmental risks. Conversely, non-violent caregiving practices, such as explaining misbehaviour or revoking privileges, correlate with higher SHAP values, suggesting a protective effect on ECD.\u003c/p\u003e \u003cp\u003eFindings support attachment theory, emphasizing that secure, nurturing environments foster emotional stability and cognitive resilience. Positive caregiving enhances neural plasticity, promoting adaptation and developmental progress (Bornstein et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Interventions should focus on reducing psychological aggression and promoting non-violent caregiving strategies to mitigate disparities in developmental outcomes and support children with functional difficulties.\u003c/p\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThis study highlights the complex factors influencing early childhood development (ECD) in rural and urban settings, using ensemble machine learning models and SHapley Additive exPlanations (SHAP) to identify and rank 15 key predictors. The findings align with Bronfenbrenner\u0026rsquo;s PPCT Bioecological Theory, emphasizing the dynamic interactions between individual and environmental factors. Child age emerged as the most influential predictor, with its impact shaped by other variables such as child functioning difficulties, caregiver education, and cognitive caregiving. Cognitive caregiving and caregiver education were significant across contexts, with maternal education amplifying cognitive stimulation efforts\u0026mdash;ranking higher in urban areas due to greater educational access.\u003c/p\u003e \u003cp\u003eRural-urban disparities were evident, with rural wealth enhancing home environments and urban overcrowding negatively affecting child development. Socioemotional caregiving, while ranking lower overall, positively interacted with cognitive caregiving in urban areas, reinforcing the link between emotional support and cognitive growth. Child functioning difficulties also influenced ECD, with developmental delays exacerbated by psychological aggression. This highlights the need for non-violent caregiving strategies to foster resilience and cognitive adaptability. Ultimately, the study calls for targeted interventions that address rural-urban gaps by improving caregiver education, promoting holistic caregiving practices, enhancing home environments, and supporting children with functional challenges. A nuanced, context-aware approach is essential to ensure equitable developmental opportunities for all children.\u003c/p\u003e \u003cp\u003eThis study possesses several notable strengths. Firstly, the application of ensemble machine learning models enhances predictive accuracy by capturing complex, non-linear relationships between multiple predictors. This approach surpasses the limitations of conventional statistical methods, offering a more nuanced understanding of early childhood development (ECD) determinants (Shukla et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Additionally, the use of SHapley Additive exPlanations (SHAP) significantly strengthens the study by providing both global and local interpretability of model outputs. This fosters transparency and aids in understanding predictor importance, bridging the gap between complex algorithms and practical application.\u003c/p\u003e \u003cp\u003eMoreover, the study's theoretical grounding in Bronfenbrenner\u0026rsquo;s PPCT Bioecological Theory of Human Development adds robustness to the findings. This alignment with established human development theories ensures that the study's conclusions are both conceptually sound and empirically supported. Another key strength lies in the study\u0026rsquo;s context-specific insights, which highlight rural-urban disparities in ECD predictors. By examining these environmental variations, the study offers tailored insights that support context-sensitive policymaking and intervention strategies (LeCroy et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDespite these strengths, several limitations warrant consideration. Firstly, \u003cb\u003edata limitations\u003c/b\u003e may have influenced the predictive accuracy of the models. While machine learning offers advanced analytic capabilities, its reliance on the quality and scope of input data means that unmeasured confounders\u0026mdash;such as parental mental health or community-level resources\u0026mdash;could affect developmental outcomes. Secondly, the study\u0026rsquo;s cross-sectional design restricts the ability to establish causality. Although SHAP values highlight predictor importance, they do not imply direct cause-and-effect relationships. Longitudinal data would better capture developmental trajectories over time, offering deeper insights into how early childhood factors interact dynamically.\u003c/p\u003e \u003cp\u003eBuilding on these strengths and limitations, several avenues for future research emerge. Firstly, adopting longitudinal designs would enable researchers to track developmental trajectories over time, offering a more dynamic understanding of how predictors influence ECD (Patel et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Longitudinal studies can illuminate the temporal relationships between variables, enhancing causal inferences.\u003c/p\u003e \u003cp\u003eMoreover, integrating multimodal data\u0026mdash;such as neuroimaging and genetic markers\u0026mdash;alongside traditional sociodemographic predictors could further enrich model accuracy. This holistic approach may uncover hidden determinants of ECD, contributing to more comprehensive models (Shukla et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eETHICS APPROVAL AND CONSENT TO PARTICIPATE\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study utilized secondary data obtained from the Multiple Indicator Cluster Survey (MICS). Access to the dataset was granted upon request, following the submission of a formal email to the MICS team, clearly stating the objectives and intended use of the data. Ethical approval was not required, as the data is publicly available and anonymized, with no identifiable information about the participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCONSENT FOR PUBLICATION\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAVAILABILITY OF DATA AND MATERIALS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets analyzed during the current study are available in the UNICEF MICS repository, accessible via https://mics.unicef.org/ upon reasonable request and registration.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCOMPETING INTERESTS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNone\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFUNDING\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAUTHORS\u0026apos; CONTRIBUTIONS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eODG conceptualized the study, conducted the data processing and analysis, interpreted the findings, and drafted the manuscript.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;JOA provided academic supervision, contributed to refining the study design and methodology, offered critical revisions to the manuscript, and approved the final version.\u003c/p\u003e\n\u003cp\u003eBMG contributed to the interpretation of findings, and provided substantial feedback on manuscript revision.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eACKNOWLEDGEMENTS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSpecial thanks the UNICEF MICS team for providing access to the dataset. Special thanks to Prof A. Adebowale, Dr. R.F. Afolabi and other faculty members for their academic guidance throughout the research.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAguilar, G. R.,\u0026amp; Sumner,A. (2019).Who are the world\u0026rsquo;s poor? A new profile of global multidimensional poverty (Working Paper 499). \u003cem\u003eCenter for Global Development\u003c/em\u003e. www.cgdev.org/sites/default/files/who-are-worlds-poor-new-profile-global-multidimensionalpoverty.pdf\u003c/li\u003e\n\u003cli\u003eAlkire, S., and Foster, J. 2011. Counting and multidimensional poverty measurement. \u003cem\u003eThe Journal of Public Economics, 95, 476\u0026ndash;487. https://doi.org/10.1016/j.jpubeco.2010.11.006\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eAzur, M. J., Stuart, E. A., Frangakis, C., and Leaf, P. J. 2011. Multiple imputation by chained equations: what is it and how does it work? \u003cem\u003eInternational Journal of Methods in Psychiatric Research\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(1), 40\u0026ndash;49. https://doi.org/10.1002/mpr.329\u003c/li\u003e\n\u003cli\u003eBast, J., and Reitsma, P. 1998.Analyzing the development of individual differences in terms of Matthew effects in reading: Results from a Dutch longitudinal study. \u003cem\u003eDevelopmental Psychology\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e(6), 1373\u0026ndash;1399. https://doi.org/10.1037/0012-1649.34.6.1373 \u003c/li\u003e\n\u003cli\u003eBizzego, A., Gabrieli, G., Lim, M., Rothenberg, W. A., Bornstein, M. H., and Esposito, G. 2022. Predictors of early childhood development: A machine learning approach. In \u003cem\u003eParenting and Child Development in Low- and Middle-Income Countries\u003c/em\u003e (pp. 210\u0026ndash;239). https://doi.org/10.4324/9781003044925-6 \u003c/li\u003e\n\u003cli\u003eBlack, M. M., Walker, S. P., Fernald, L. C. H., Andersen, C. T., DiGirolamo, A. M., Lu, C., McCoy, D. C., Fink, G., Shawar, Y. R., Shiffman, J., Devercelli, A. E., Wodon, Q. T., Vargas-Bar\u0026oacute;n, E., and Grantham-McGregor, S. 2016. Early childhood development coming of age: science through the life course. \u003cem\u003eThe Lancet\u003c/em\u003e, \u003cem\u003e389\u003c/em\u003e(10064), 77\u0026ndash;90. https://doi.org/10.1016/s0140-6736(16)31389-7 \u003c/li\u003e\n\u003cli\u003eBlack, M. M., Walker, S. P., Fernald, L. C., Andersen, C. T., DiGirolamo, A. M., Lu, C., McCoy, D. C., Fink, G., Shawar, Y. R., Shiffman, J., Devercelli, A. E., Wodon, Q. T., Vargas-Bar\u0026oacute;n, E., Grantham-McGregor, S., and Lancet Early Childhood Development Series Steering Committee. 2017. Early childhood development coming of age: Science through the life course. \u003cem\u003eThe Lancet\u003c/em\u003e, \u003cem\u003e389\u003c/em\u003e(10064), 77\u0026ndash;90. https://doi.org/10.1016/S0140-6736(16)31389-7 \u003c/li\u003e\n\u003cli\u003eBornstein, M. H., and Leventhal,T. (Eds.). 2015. \u003cem\u003eEcological settings and processes in developmental systems.Volume 4 of the handbook of child psychology and developmental science \u003c/em\u003e(Editor-in- Chief: R. M. Lerner, 7th ed.).Wiley. \u003c/li\u003e\n\u003cli\u003eBornstein, M. H., Bradley, R. H., Rothenberg, W. A., Deater-Deckard, K., Esposito, G., Lansford, J. E., Zietz, S., Putnick, D. L., and Bizzego, A. 2022. Parenting and child development in Low- and Middle-Income countries. In \u003cem\u003eRoutledge eBooks\u003c/em\u003e. https://doi.org/10.4324/9781003044925 \u003c/li\u003e\n\u003cli\u003eBradley, R. H. 2020. The child\u0026rsquo;s environment. In Part of a book series: M. H. Bornstein (Ed.), \u003cem\u003eElements in child development\u003c/em\u003e. Cambridge University Press. https://doi.org/10.1017/9781108866040 \u003c/li\u003e\n\u003cli\u003eBradley, R. H. 2020. The child\u0026rsquo;s environment. In Part of a book series: M. H. Bornstein (Ed.), \u003cem\u003eElements in child development\u003c/em\u003e. Cambridge University Press. https://doi.org/10.1017/9781108866040 \u003c/li\u003e\n\u003cli\u003eBritto, P. R., Lye, S. J., Proulx, K., Yousafzai, A. K., Matthews, S. G., Vaivada, T., Perez-Escamilla, R., Rao, N., Ip, P., Fernald, L. C. H., MacMillan, H., Hanson, M., Wachs, T. D., Yao, H., Yoshikawa, H., Cerezo, A., Leckman, J. F., Bhutta, Z. A., \u0026amp; Early Childhood Development Interventions Review Group, for the Lancet Early Childhood Development Series Steering Committee (2017). Nurturing care: promoting early childhood development. \u003cem\u003eLancet (London, England)\u003c/em\u003e, \u003cem\u003e389\u003c/em\u003e(10064), 91\u0026ndash;102. https://doi.org/10.1016/S0140-6736(16)31390-3\u003c/li\u003e\n\u003cli\u003eBronfenbrenner, U., and Morris, P. A. 2006.The bioecological model of human development. In R. M. Lerner (Ed.),Theoretical models of human development.Volume 1 of the handbook of child psychology (Editors-in-Chief:W. Damon and R. M. Lerner, 6th ed., pp. 793\u0026ndash;828).Wiley. https://doi.org/10.1002/9780470147658.chpsy0114\u003c/li\u003e\n\u003cli\u003eDe Temple, J., and Snow, C. E. 2003. Learning words from books. In A. van Kleeck, S. A. Stahl, and E. B. Bauer (Eds.), \u003cem\u003eOn reading books to children \u003c/em\u003e(pp. 16\u0026ndash;36). Erlbaum. \u003c/li\u003e\n\u003cli\u003eErickson, Nick, Mueller J, Shirkov A., Zhang H., Larroy P ., Li M., and Smola A. 2020 \u0026quot;AutoGluon-Tabular: Robust and Accurate AutoML for Structured Data.\u0026quot; arXiv preprint arXiv:2003.06505 \u003c/li\u003e\n\u003cli\u003eGalasso, E., and Wagstaff, A. 2018. The aggregate income losses from childhood stunting and the returns to a nutrition intervention aimed at reducing stunting. \u003cem\u003eEconomics and Human Biology\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e, 225\u0026ndash;238. https://doi.org/10.1016/j.ehb.2019.01.010 \u003c/li\u003e\n\u003cli\u003eGinsburg, K. R. (2007). The importance of play in promoting healthy child development and maintaining strong Parent-Child bonds. \u003cem\u003ePEDIATRICS\u003c/em\u003e, \u003cem\u003e119\u003c/em\u003e(1), 182\u0026ndash;191. https://doi.org/10.1542/peds.2006-2697\u003c/li\u003e\n\u003cli\u003eHeckman, J. J. 2006. Skill formation and the economics of investing in disadvantaged children. \u003cem\u003eScience\u003c/em\u003e, \u003cem\u003e312\u003c/em\u003e(5782), 1900-1902.\u003c/li\u003e\n\u003cli\u003eJones, S. M., and Bouffard, S. M. 2012. Social and emotional learning in schools: From programs to strategies. \u003cem\u003eSocial Policy Report\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e(4), 1-33.\u003c/li\u003e\n\u003cli\u003eJones, S. M., and Bouffard, S. M. 2012. Social and emotional learning in schools: From programs to strategies. \u003cem\u003eSocial Policy Report\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e(4), 1-33.\u003c/li\u003e\n\u003cli\u003eLange, G., Wodon, Q., and Carey, K. 2017. The Changing Wealth of Nations 2018: Building a Sustainable Future. In \u003cem\u003eWashington, DC: World Bank eBooks\u003c/em\u003e. https://doi.org/10.1596/978-1-4648-1046-6 \u003c/li\u003e\n\u003cli\u003eLeCroy, M., Kim, R. S., Stevens, J., Hanna, D., \u0026amp; Isasi, C. (2021). Identifying key determinants of childhood obesity: A narrative review of machine learning studies. \u003cem\u003eChildhood Obesity.\u003c/em\u003e https://doi.org/10.1089/chi.2020.0324\u003c/li\u003e\n\u003cli\u003ePatel, D., Hall, G., Broadhurst, D. I., Smith, A., Schultz, A., \u0026amp; Foong, R. (2021). Does machine learning have a role in the prediction of asthma in children? \u003cem\u003ePaediatric Respiratory Reviews.\u003c/em\u003e https://doi.org/10.1016/j.prrv.2021.06.002\u003c/li\u003e\n\u003cli\u003eRaver, C. C., Smith-Donald, R., Hayes, T., and Jones, S. M. 2007. Self-regulation across different contexts: Implications for learning and early academic success. \u003cem\u003eEarly Education and Development\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(2), 303-320.\u003c/li\u003e\n\u003cli\u003eRichter, L. M., Daelmans, B., Lombardi, J., Heymann, J., Boo, F. L., Behrman, J. R., Lu, C., Lucas, J. E., Perez-Escamilla, R., Dua, T., Bhutta, Z. A., Stenberg, K., Gertler, P., Darmstadt, G. L., and Paper 3 Working Group and the Lancet Early Childhood Development Series Steering Committee 2017. Investing in the foundation of sustainable development: pathways to scale up for early childhood development. \u003cem\u003eLancet (London, England)\u003c/em\u003e, \u003cem\u003e389\u003c/em\u003e(10064), 103\u0026ndash;118. https://doi.org/10.1016/S0140-6736(16)31698-1\u003c/li\u003e\n\u003cli\u003eShonkoff, J. P., and Phillips, D. A. 2000. \u0026quot;From neurons to neighborhoods: The science of early childhood development.\u0026quot; National Academy Press.\u003c/li\u003e\n\u003cli\u003eShukla, P. K., Jain, S., \u0026amp; Kalra, S. (2022). Machine learning in childhood asthma prediction: A systematic review. \u003cem\u003e2022 10th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO)\u003c/em\u003e, 1\u0026ndash;6. https://doi.org/10.1109/ICRITO56286.2022.9964849\u003c/li\u003e\n\u003cli\u003eSpencer, N., Raman, S., O\u0026apos;Hare, B., and Tamburlini, G. 2019. Addressing inequities in child health and development: towards social justice. \u003cem\u003eBMJ paediatrics open\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e(1), e000503. https://doi.org/10.1136/bmjpo-2019-000503\u003c/li\u003e\n\u003cli\u003eStanovich, K. E. 1986. Matthew effects in reading: Some consequences of individual differences in the acquisition of literacy. \u003cem\u003eReading Research Quarterly\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(4), 360\u0026ndash;407. https://doi.org/10.1598/RRQ.21.4.1 \u003c/li\u003e\n\u003cli\u003eUnited Nations Children\u0026rsquo;s Fund 2023. The Early Childhood Development Index 2030: A new measure of early childhood development, UNICEF, New York.\u003c/li\u003e\n\u003cli\u003eWalker, S. P., Wachs, T. D., Grantham-McGregor, S., et al. 2011. Inequality in early childhood: Risk and protective factors for early child development. \u003cem\u003eThe Lancet\u003c/em\u003e, \u003cem\u003e378\u003c/em\u003e(9799), 1325-1338.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization 2012. Early childhood development and disability: A discussion paper. \u003c/li\u003e\n\u003cli\u003eWorld Health Organization. 2021. \u003cem\u003eWorld health statistics. Monitoring health for the SDGs\u003c/em\u003e. World Health Organization.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"bmc-public-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pubh","sideBox":"Learn more about [BMC Public Health](http://bmcpublichealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pubh/default.aspx","title":"BMC Public Health","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Early childhood development, interpretable ensemble models, machine learning, caregiving practices, SDG Target 4.2. Count: 244 words","lastPublishedDoi":"10.21203/rs.3.rs-8615011/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8615011/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eEarly childhood development (ECD) is essential for lifelong health, education, and economic productivity, as emphasized by SDG Target 4.2, which seeks universal access to quality pre-primary education by 2030. In Nigeria, ECD are significantly impacted by poverty, malnutrition, and systemic inequities. However, there is limited research on context-specific determinants such as caregiving practices, parental discipline, and home environments. This study developed a Multilayer Stacked Weighted Ensemble Model (MSWEM) and analyse data from the 2021 Nigeria Multiple Indicator Cluster Survey (MICS), comprising 15,899 children aged 24\u0026ndash;59 months. Statistical techniques included descriptive analysis, binary logistic regression for predictor identification and machine learning for advanced prediction and SHAP (SHapley Additive exPlanations) analysis for model interpretability. The model development was implemented in Python 3.7, integrating ensemble methods to enhance prediction accuracy\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe MSWEM achieved a high predictive accuracy with an ROC-AUC of 93% and accuracy of 85.2%. The model correctly predicted 46.7% on \u0026ndash; track children in urban areas and 24.7% on-track children in rural areas. Factors influencing the likelihood of a child being on track include age of the child, cognitive caregiving practice, overall physical home environment, mother\u0026rsquo;s education, age of woman, to mention but a few\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThis study highlights the utility of interpretable machine learning in understanding and predicting ECD outcomes. These efforts are essential for ensuring equitable opportunities for all children in Nigeria, aligning with national development goals and global progress toward SDG 4.2.\u003c/p\u003e","manuscriptTitle":"Rural-Urban Disparities in Early Childhood Development: Insights from Interpretable Ensemble Machine Learning Model","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-02 05:28:10","doi":"10.21203/rs.3.rs-8615011/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-03-19T08:09:52+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"93964616138475640974527004576023222599","date":"2026-03-13T21:36:17+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"1345968173712970648575623904700090152","date":"2026-03-12T09:07:24+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-12T02:33:16+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"89176578126473027081343446821185411754","date":"2026-03-06T20:55:38+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-01-28T10:48:43+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-01-21T08:45:51+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-01-20T11:24:56+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-20T11:21:22+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Public Health","date":"2026-01-16T03:35:05+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-public-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pubh","sideBox":"Learn more about [BMC Public Health](http://bmcpublichealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pubh/default.aspx","title":"BMC Public Health","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"760ab406-83dd-48e0-a54e-40c293aeb5f7","owner":[],"postedDate":"February 2nd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-02-02T05:28:10+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-02 05:28:10","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8615011","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8615011","identity":"rs-8615011","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}