A Machine Learning and Explainable AI-Based Multiclass Police Fraud Prediction Scheme with SHAP Based Interpretability

By resulting in monetary loss, public mistrust, psychological distress, and missed opportunities, police fraud can seriously impair citizens. A machine learning-based multiclass fraud prediction system with high accuracy that takes into consideration financial fraud, impersonation fraud, general duty fraud, and no fraud category was not developed in earlier studies. They failed to incorporate appropriate preprocessing, feature selection, annotation, class imbalance problem solving, hyper parameter tweaking, and data set preparation techniques. To solve the existing limitations, this paper delivers an ML based multi class police fraud prediction system with advanced data preprocessing, SMOTE based class imbalance handling, ensemble technique based feature selection, grid search CV based hyper paramer tuning, and SHAP base model interpretability. Because of its maximum accuracy of 96.66% and F1 score of 96.01%, the XGBoost model is chosen as the best machine learning model for police fraud prediction. According to the comparison results, the suggested XGBoost scheme outperforms the current works by achieving 2.6% gain in accuracy and 3% gain in F1 score.