KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 42 2026 02 10 Explainable Machine Learning Prediction of Dropout Risk Using Psychosocial and Cognitive Variables Explainable Machine Learning Prediction of Dropout Risk Using Psychosocial and Cognitive Variables 1 11 10.61838/kman.jayps.5089 EN Journal Article 2025 09 23 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> The present study aimed to develop and validate an explainable machine learning model capable of accurately predicting school dropout risk by integrating psychosocial and cognitive variables.</p> <p><strong>Methods and Materials: </strong>A cross-sectional predictive design was employed with a sample of 1.172 secondary school students from three federal states in Germany. Standardized instruments were used to assess psychosocial variables including depressive symptoms, academic self-efficacy, school belonging, teacher and peer support, self-regulation, and academic motivation. Cognitive performance was measured through computerized tasks assessing working memory, processing speed, and fluid reasoning. Socioeconomic status and migration background were included as contextual covariates. Data preprocessing involved multiple imputation, normalization, and class imbalance correction using SMOTE. The dataset was partitioned into training, validation, and independent test subsets. Multiple supervised learning algorithms—logistic regression with elastic net regularization, support vector machines, random forest, and gradient boosting—were trained and compared using cross-validated hyperparameter optimization. Model performance was evaluated using AUC, balanced accuracy, F1-score, and calibration indices. Explainability was ensured through SHAP-based global and local feature attribution analyses.</p> <p><strong>Findings: </strong>Gradient boosting achieved the highest predictive performance (AUC = 0.92; balanced accuracy = 0.86), significantly outperforming linear models. Psychosocial variables demonstrated stronger predictive power than cognitive variables alone, yet the integration of both domains significantly improved overall model accuracy. Depressive symptoms, academic self-efficacy, and school belonging emerged as the most influential predictors, while processing speed and working memory provided incremental predictive validity. Nonlinear threshold effects were observed, indicating that elevated emotional distress and reduced cognitive efficiency substantially increased dropout probability.</p> <p><strong>Conclusion: </strong>The findings demonstrate that explainable machine learning models integrating psychosocial and cognitive indicators can reliably predict dropout risk while preserving interpretability.</p> </td> </tr> </tbody> </table> https://www.journals.kmanpub.com/index.php/jayps/article/download/5089/9180