KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 49 2026 09 10 Predicting High School Dropout Risk from Academic Engagement, Attendance, Self-Efficacy, and Family Support Using TabNet Predicting High School Dropout Risk from Academic Engagement, Attendance, Self-Efficacy, and Family Support Using TabNet 1 20 10.61838/kman.jayps.5684 EN Journal Article 2025 12 14 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> This study aimed to develop and evaluate an interpretable TabNet model for predicting verified high school dropout among French upper-secondary students using academic engagement, attendance, academic self-efficacy, and perceived family support.</p> <p><strong>Methods and Materials:</strong> A prospective longitudinal predictive design was employed with 1,248 students aged 15–18 years recruited from 12 public high schools across four regions of France. Baseline data on behavioral, emotional, cognitive, and overall academic engagement, academic self-efficacy, and perceived family support were collected through validated French-language questionnaires, while attendance, excused and unexcused absences, lateness, and dropout status were obtained from school administrative records. Students were followed through the end of the academic year and subsequent enrollment verification. Data from nine schools were used for model development, including training and validation subsets, while three schools formed an independent test set. TabNet performance was compared with penalized logistic regression, random forest, and gradient-boosted trees using discrimination, calibration, and classification indices.</p> <p><strong>Findings:</strong> Students who dropped out had significantly lower overall, behavioral, emotional, and cognitive engagement, lower attendance, weaker academic self-efficacy, and lower perceived family support, alongside significantly higher excused absences, unexcused absences, and lateness, with all comparisons reaching p &lt; .001. In the independent test set, TabNet achieved an AUROC of .934, an AUPRC of .612, sensitivity of .857, specificity of .919, balanced accuracy of .888, and an F1 score of .640. TabNet significantly outperformed penalized logistic regression and random forest in AUROC, while its advantage over gradient-boosted trees was not statistically significant. Unexcused absence, overall attendance, behavioral engagement, academic self-efficacy, and family support were the most influential predictors.</p> <p><strong>Conclusion:</strong> TabNet accurately predicted high school dropout by integrating behavioral, academic, and relational indicators and may support early, targeted, and interpretable school-based prevention when used alongside professional judgment.</p> </td> </tr> </tbody> </table> https://journals.kmanpub.com/index.php/jayps/article/download/5684/11788