<![CDATA[Drug abuse among adolescents is a highly complex social and public health problem, influenced by multifactorial interactions between social, demographic, psychological, and environmental factors. Current prevention efforts are generally generalist and educational in nature, thus significantly limiting their ability to identify individuals with specific risk vulnerabilities early. Furthermore, existing predictive models often rely on costly and time-consuming clinical measurement instruments, making them difficult to implement on a large scale in schools. To address these challenges, this study aims to develop and evaluate a predictive model for the risk of drug abuse in adolescents using the Random Forest algorithm. This approach focuses on utilizing social, demographic, and environmental variables in a more practical and accessible manner. The dataset used comprises 10,000 respondents, with methodological steps including data preprocessing, feature engineering for classification of three risk classes (Low, Medium, High) using a quantile approach, and dividing the dataset into 80% training data and 20% testing data. The model was comprehensively evaluated using accuracy, precision, recall, F1-score, and confusion matrix metrics. The results showed that the Random Forest model achieved an accuracy rate of 33% in the multi-class classification scheme, and significantly increased to 51.95% in the binary classification scheme. Furthermore, feature importance analysis revealed that smoking prevalence, peer influence, and family background conditions were the predictors with the most dominant contribution in mapping adolescent vulnerability. The main contribution of this study is the design of a more granular risk categorization scheme and providing an initial foundation for the development of an adaptive, practical, and highly scalable machine learning-based early warning system to support preventive interventions in educational institutions.]]>
