Answers are grounded in this paper and cite their sources.
Drug misuse is a growing global health issue, particularly among university students, leading to self-harm and severe social consequences. Addicted youth often exhibit distinct behavioral, social, and familial patterns, underscoring the need for reliable early detection. This study proposed an explainable ensemble framework (method-4) for detecting student drug addiction (SDA) which is based on novel 1D CNN accompanied by a full linked system (FLSs) (method-1), and 1D CNN–LSTM–FLS (method-2). Both Method-1 and Method-2 were enhanced through the incorporation of a unit of feature layers (UFLs) to improve feature representation. The methods were trained on 12,745 student records from university and rehabilitation centers, incorporating academic, social, financial, health, and legal features. The proposed framework (method-4) secured a specificity of 88.62% and a log loss of 9.30%, alongside an overall accuracy of 74.00%, precision of 83.92%, and F1-score of 69.55% on the dataset 1, outperforming competitive machine learning and deep learning baselines. Furthermore, independent evaluation on a secondary cohort verified robust cross-domain generalizability with 71.35% accuracy. Specifically, post-hoc SHAP attribution and a variance-based feature importance metric identified academic decline, legal infractions, and social isolation as primary risk determinants, offering actionable, transparent decision support for proactive campus and rehabilitation interventions. The code is publicly available at here https://github.
No takes yet. Share an insight, caveat, or question.
Saha et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: