Enhancing Insurance Fraud Detection Accuracy with Integrated Machine Learning and Statistical Methods

Abstract The insurance industry plays a critical role in managing risks and providing financial security globally. However, the industry faces challenges, particularly with the increasing complexity of fraudulent activities. To address these challenges, this work seeks to construct suitable decision models by integrating methods such as feature discretization, feature selection, data resampling, and binary classification, in order to create a prediction system for identifying insurance fraud. The research investigates various scenarios, including different combinations of classifiers, feature selection methods, feature discretization techniques, and data resampling strategies, and the performance of the predictive system is evaluated using established metrics. The experimental results revealed that integrating multiple methodologies during data preprocessing significantly enhances the performance of classification models. The model that utilizes the KBD + RFE + Over + RF scenario achieves the highest AUC and F1-score, indicating exceptional performance in detecting insurance fraud. Our research demonstrates that the proposed models’ ability to predict insurance fraud has been significantly enhanced by utilizing resampling methods and highlights the importance of these techniques in improving the efficiency of the utilized integrated artificial intelligence techniques. In addition, the article concludes that the insurance industry can greatly benefit from modern predictive methods to make sound decisions.