An Automated System for Detecting Property Insurance Fraud Using Machine Learning

Detecting property insurance fraud is critical for reducing financial losses and ensuring fair claim processing. Traditional methods of detecting insurance fraud had several drawbacks, including no feature selection process, no hyper parameter tuning, lower accuracy, and class imbalance problems. To address the aforementioned shortcomings, this paper examines advanced ML (machine learning) techniques for accurately detecting property insurance fraud. To determine the best model for predicting fraudulent activities, this paper tested several machine learning models, including Gradient Boosting, classical ML classifiers, and Stacking Ensemble methods. To address class imbalance and improve model performance, the selected model incorporates proper feature selection, hyper parameter tuning, and SMOTE techniques (synthetic minority over-sampling). The Stacking Ensemble method outperformed the other ML models, achieving an accuracy of 96% and a recall of 94%. The experimental results show that the proposed stacking ensemble-based prediction scheme improves accuracy by 3.4% and recall by 2.7% over previous works. This article also includes a web application for assisting with property insurance fraud, which includes ML-based fraud prediction, question submission, answer checking, and blog post access. According to the findings, more than 54% of users expressed satisfaction with the web application's usefulness for detecting property fraud.