Prediction of juvenile crime in Bangladesh due to drug addiction using machine learning and explainable AI techniques

In this paper, we depict the concept of predicting juvenile crime due to drug addiction using machine learning techniques. With the exponentially increasing crime rate, it has become a daunting task to prevent them at an evenly faster rate. Therefore, we have worked with several variables that are inextricably linked with adolescent offenses. The study identifies the causes of adolescent drug addiction by pinpointing the behavioral disorders and predicts their tendency of crime engagement. Using a dataset of both drug-addicted and non-addicted teens, the proposed approach predicts the latent connections to future under-aged crimes. The dataset was collected from four drug rehabilitation centers and high school students. To achieve better prediction, we have used multiple machine learning techniques such as Logistic Regression, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Random Forest, XGBoost, and Multilayer Perceptrons (MLPs). We have evaluated several performance parameters, namely accuracy, precision, recall, and f1 score, and compared them in light of individual algorithms. The experimental results showed that MLPs algorithm outperformed others attaining 95.36% accuracy, whereas other algorithms achieved accuracies ranging from 89 to 94%. We have also extracted the best features by applying Univariate Feature Selection method and Recursive Feature Elimination (RFE) method. The proffered estimation was explained by Shapley Additive explanation (SHAP) and Local Interpretable Model-Agnostic Explanation (LIME) techniques. The proposed method can allow counter-measures of the juvenile criminal activities induced from addiction by producing explained crime forecast.