A Classification of Drug Types Based on Crime Modus Operandi Using Machine Learning

This study adopts a judgment-based approach to classify drug types based on their associated criminal modus operandi. A total of 105 drug transaction cases, extracted from legal judgments, were analyzed. Each case includes key variables such as the method of contact, payment method, transaction location, delivery method, and the specific type of drug involved. Three machine learning models—Random Forest, XGBoost, and Support Vector Machine (SVM)—were employed to classify drug types according to crime methods. Model performance was evaluated using accuracy, precision, recall, and F1-score. Additionally, SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) were used to interpret the influence of each crime method on the model's predictions. The results demonstrated that all three machine learning models exhibited strong predictive performance, with Random Forest and XGBoost achieving particularly high accuracy. These models identified the methods of contact and delivery as the most influential predictors. However, the SVM model, while comparable in predictive accuracy, showed limitations in its ability to interpret key variables effectively.