Psychological Research in Individuals with Exceptional Needs

The present study aimed to develop and evaluate an explainable machine learning model for predicting vocational success among young adults with intellectual disabilities based on self-determination, career adaptability, and family empowerment. This cross-sectional predictive study was conducted among 428 young adults with mild to moderate intellectual disabilities recruited from vocational rehabilitation centers, supported employment programs, and transition-to-work services across Canada. Participants completed standardized measures of vocational success, self-determination, career adaptability, and family empowerment. Data preprocessing procedures included standardization, missing-value imputation, and quality screening. Several machine learning algorithms, including Linear Regression, Support Vector Regression, Random Forest, Gradient Boosting, and Extreme Gradient Boosting (XGBoost), were trained and evaluated using a 70:30 training-testing split and five-fold cross-validation. Model performance was assessed using the coefficient of determination (R²), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). SHapley Additive exPlanations (SHAP) analysis was employed to identify feature importance and improve interpretability of model predictions. Significant positive associations were observed between vocational success and self-determination (r = .68, p < .001), career adaptability (r = .61, p < .001), and family empowerment (r = .54, p < .001). Among the predictive models tested, XGBoost demonstrated the highest predictive accuracy, accounting for 76% of the variance in vocational success (R² = .76), with the lowest prediction errors (RMSE = 6.94, MAE = 5.13, MAPE = 6.95%). SHAP analysis revealed that self-determination was the most influential predictor, contributing 41.5% of total model importance, followed by career adaptability (30.4%) and family empowerment (22.7%). Collectively, these three variables accounted for more than 94% of the predictive contribution within the final model. The findings indicate that vocational success among young adults with intellectual disabilities can be predicted with high accuracy using machine learning techniques. Self-determination emerged as the strongest predictor, followed by career adaptability and family empowerment, highlighting the importance of both personal agency and supportive family environments. The results support the use of explainable artificial intelligence approaches in disability and vocational research and suggest that interventions targeting these psychosocial factors may enhance employment outcomes and long-term vocational success.

The present study aimed to develop and interpret an explainable artificial intelligence model of social participation in children with intellectual disabilities by examining the predictive contributions of adaptive behavior, communication competence, and parental involvement using a SHAP-based CatBoost algorithm. This cross-sectional predictive modeling study was conducted among 412 children with mild to moderate intellectual disabilities aged 8–16 years who were recruited from special education schools, rehabilitation centers, and inclusive educational settings across Hungary. Social participation was assessed using the Participation and Environment Measure for Children and Youth (PEM-CY), while adaptive behavior, communication competence, and parental involvement were measured using standardized psychometric instruments. Following data preprocessing and screening procedures, a CatBoost machine learning model was developed to predict social participation outcomes. Model performance was evaluated using the coefficient of determination (R²), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). To enhance interpretability, SHapley Additive exPlanations (SHAP) analyses were conducted to determine the relative importance and direction of predictor contributions. Pearson correlation analyses demonstrated significant positive associations between social participation and adaptive behavior (r = .781, p < .001), communication competence (r = .724, p < .001), and parental involvement (r = .647, p < .001). The CatBoost model exhibited excellent predictive performance, explaining 89.2% of the variance in social participation within the test dataset (R² = .892), with RMSE = 4.97, MAE = 3.89, and MAPE = 5.84%. SHAP analyses identified adaptive behavior as the strongest predictor of social participation (42.7% relative importance), followed by communication competence (32.8%) and parental involvement (19.7%). Together, these three variables accounted for more than 95% of the model’s total explanatory influence. Furthermore, SHAP dependence analyses revealed nonlinear interactions, indicating that the positive effects of adaptive behavior became increasingly pronounced at higher levels of communication competence and parental involvement. The findings demonstrate that social participation among children with intellectual disabilities is primarily influenced by adaptive behavior, communication competence, and parental involvement. The SHAP-based CatBoost model provided highly accurate and interpretable predictions, highlighting the value of explainable artificial intelligence for understanding participation outcomes in developmental disability research. These results support the development of family-centered and skills-focused interventions aimed at enhancing social inclusion and community participation among children with intellectual disabilities.

The present study aimed to investigate the predictive contributions of sensory sensitivity, parent–child attachment, and social competence to emotional well-being among children with Autism Spectrum Disorder (ASD) using a Light Gradient Boosting Machine (LightGBM) model and explainable artificial intelligence techniques. This cross-sectional predictive study was conducted among 428 children diagnosed with Autism Spectrum Disorder in South Africa. Emotional well-being served as the target variable, while sensory sensitivity, parent–child attachment, and social competence were entered as predictor variables. Data were collected using standardized caregiver-report instruments with established psychometric properties. After preliminary data screening and preprocessing, the dataset was randomly divided into training and testing subsets using an 80:20 ratio. A LightGBM algorithm was implemented to develop the predictive model, and hyperparameter optimization was performed through five-fold cross-validation. Model performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and mean squared error (MSE). Feature importance analysis and Shapley Additive Explanations (SHAP) were employed to determine the relative contribution and direction of influence of each predictor variable. The LightGBM model demonstrated strong predictive performance, explaining 82.1% of the variance in emotional well-being within the testing dataset (R² = 0.821). Correlation analyses revealed significant positive associations between emotional well-being and both parent–child attachment (r = 0.68, p < .01) and social competence (r = 0.74, p < .01), whereas sensory sensitivity was significantly negatively associated with emotional well-being (r = −0.61, p < .01). Feature importance analysis indicated that social competence was the most influential predictor (39.8%), followed by parent–child attachment (33.3%) and sensory sensitivity (26.9%). SHAP analyses confirmed these findings, demonstrating that higher levels of social competence and stronger parent–child attachment increased predicted emotional well-being, whereas elevated sensory sensitivity reduced emotional well-being predictions. The findings indicate that emotional well-being among children with ASD is strongly shaped by social, relational, and sensory factors. Social competence emerged as the most influential predictor, highlighting the importance of adaptive social functioning for positive emotional outcomes. Secure parent–child attachment also contributed substantially to emotional well-being, while sensory sensitivity functioned as a significant risk factor. The high predictive accuracy of the LightGBM model demonstrates the value of machine learning approaches for identifying key determinants of emotional well-being and supports the development of targeted interventions focusing on social skills enhancement, family relationships, and sensory regulation to improve psychological outcomes among children with ASD.

The present study aimed to develop and evaluate a CatBoost machine learning model for predicting social anxiety among students with specific learning disabilities based on rejection sensitivity, self-compassion, and peer victimization, while determining the relative contribution of each predictor to model performance. This cross-sectional predictive study was conducted among 428 students with specific learning disabilities enrolled in secondary schools in Santiago, Chile. Participants completed standardized measures assessing social anxiety, rejection sensitivity, self-compassion, and peer victimization. Following data preprocessing and preliminary statistical analyses, a CatBoost machine learning algorithm was implemented to predict social anxiety levels. The dataset was divided into training (80%) and testing (20%) subsets, and hyperparameter optimization was performed using five-fold cross-validation. Model performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Feature importance analysis and SHAP (Shapley Additive Explanations) values were calculated to identify the relative influence and directional effects of the predictors. The CatBoost model demonstrated strong predictive performance, accounting for 81.2% of the variance in social anxiety scores within the testing dataset (R² = .812). The model yielded low prediction errors (RMSE = 5.08, MAE = 3.94, MAPE = 8.46%), indicating high accuracy and generalizability. Feature importance analysis revealed that rejection sensitivity was the strongest predictor of social anxiety (41.82%), followed by peer victimization (34.67%) and self-compassion (23.51%). SHAP analyses showed that higher levels of rejection sensitivity and peer victimization were associated with increased social anxiety, whereas higher self-compassion was associated with lower social anxiety. Correlational analyses further indicated significant positive associations between social anxiety, rejection sensitivity, and peer victimization, alongside a significant negative association between social anxiety and self-compassion. The findings demonstrate that social anxiety among students with specific learning disabilities is strongly influenced by interpersonal vulnerability factors and psychological resilience resources. Rejection sensitivity and peer victimization represent significant risk factors, whereas self-compassion serves as an important protective factor. CatBoost modeling provides an effective and interpretable approach for identifying students at risk and may support the development of targeted school-based prevention and intervention programs aimed at improving emotional well-being and social functioning.