Enhancing Risk Assessment in Auto Insurance with Data-Driven Insights using Machine Learning

Abstract

The insurance firms, detecting auto insurance fraud is a major difficulty that may result in large financial losses. Insurance customers suffer substantial monetary losses, including higher premiums, as a result of claims that are forged. Manual inspections and rule-based techniques are the foundation of conventional fraud detection techniques, which are ineffective and unable to keep up with changing fraud trends. Machine learning (ML) is used in this investigation technique to improve fraud detection accuracy by analyzing vehicle insurance claim data. Extensive data preprocessing was applied, including handling missing values, feature selection, one-hot encoding, Min-Max normalization, and oversampling to address the severe class imbalance. A Random Forest (RF) classifier accomplished the uppermost accuracy (97.5%), outperforming Logistic Regression (LR) (87.1%) and EXtreme Gradient Boosting (XGBoost) (77.61%). Random Forest (RF) also showed superior precision (95.6%), recall (99.5%), and F1-score (97.5%), with an AUC of 0.98 from the ROC analysis, confirming its effectiveness. Despite its strong performance, limitations include dataset age and synthetic data from oversampling. The proposed approach offers an automated, scalable, and efficient fraud detection system, enhancing decision-making in the business of protection. Using machine learning (ML), this research offers an inexpensive and effective way to improve automobile insurance fraud detection.

Keywords: Auto Insurance, Risk Assessment, Insurance Fraud Prediction, Fraud Detection, Machine Learning (ML), vehicle insurance data