According to the unpredictable as well as lethal nature of drugs, forecasting psychoactive drug consumption poses an enormous challenge to law enforcement authorities tasked with maintaining public safety. This study aims to address this issue by leveraging machine learning techniques to predict drug usage among the population of Bangladesh. Data was collected from 1,938 participants via a Google form, encompassing a wide range of demographic and behavioral information. We employed a variety of supervised learning models, including Gaussian Naive Bayes, Support Vector Machines (SVM), Random Forest, Decision Trees, Ada Boost, Ridge Classifier, Logistic Regression, and a Voting Classifier. Our analysis focused on ten different categories of drugs—alcohol, nicotine, mushrooms, chocolate, heroin, cocaine, benzodiazepines, caffeine, LSD, and ecstasy—performing separate classifications for each. Notably, the prediction model for chocolate consumption demonstrated an exceptional accuracy rate of 98.97%, utilizing a combination of Logistic Regression and SVM. Additionally, to enhance the interpretability of our predictive models, we incorporated the Local Interpretable Model-Agnostic Explanations (LIME) technique.