INSURANCE FRAUD DETECTION USING MACHINE LEARNING ON CLASSIMBALANCED DATASETS WITH MISSING VALUES

Abstract

Insurance fraud is a major problem for insurers, especially in the vehicle insurance industry. It affects pricing tactics and causes financial losses. Class imbalance, when fraudulent claims are far less common than legitimate claims, frequently affects fraud detection models, and missing data makes the task even more difficult. Two vehicle insurance datasets—a conventional dataset and an Egyptian real-life dataset—are used in this study to address these problems. In order to improve the accuracy and prediction potential of the model, the AdaBoost Classifier is incorporated into the suggested methodology, which also addresses missing data and class imbalance. The findings show that correcting class imbalance is essential to enhancing model performance, and handling missing data also helps to produce predictions that are more trustworthy. By increasing prediction accuracy and decreasing overfitting, which is frequently a problem in fraud detection models, the AdaBoost Classifier greatly outperforms current methods. This study offers insightful information about how enhancing data quality and utilising cutting-edge algorithms like AdaBoost can improve fraud detection systems, ultimately resulting in more successful identification of fraudulent claims. These improvements can greatly help insurance businesses make better decisions, lower financial losses, and improve pricing strategies.