Optimizing Predictive Models for Drug-Induced Suicidal Risk: A Data-Balanced Framework with Explainable AI

A thorough investigation is needed to understand the rising global patterns of drug addiction and suicide rates. This study explores the link between drug addiction and suicidal ideation to develop a prediction model for early detection and support of at-risk individuals. An exhaustive analysis employs computational complexity, performance metrics, AUC-ROC curves, and cross-entropy loss to meticulously assess and compare the efficacy by using a suite of machine learning algorithms, such as Logistic Regression, Random Forest, Gradient Boosting, K-Nearest Neighbors (KNN), and Extra Trees Classifiers on imbalanced and balanced datasets. By implementing advanced techniques like SMOTE for data balancing, the reliability and precision of predictions are ensured. The Extra Trees Classifier, with hyper-parameter tuning, is the most effective model, achieving a 94% accuracy rate, 0.95 AUC score, and 0.2589 cross-entropy loss. The study emphasizes explainable AI, providing logical insights and detailed explanations for model selection across different parameters. This thorough analysis ensures transparency and fosters trust in the predictive capabilities of models. The findings set new benchmarks in predictive modeling and intervention strategies, with implications for clinical practices and public health strategies. The potential to transform intervention methods and save lives underscores the critical importance of this study.