An Explainable Intelligent System for Auto Insurance Fraud Detection Using Naïve Bayes

Abstract - Insurance fraud detection represents a persistent challenge that significantly impacts both insurance companies and policyholders through increased premiums and operational costs. This research presents a comprehensive web-based fraud detection solution developed using the C# ASP.NET framework integrated with Naive Bayes classification algorithms. The proposed system implements a multi-tiered user access structure comprising four distinct roles: administrators who oversee city-wide and branch operations while managing user account creation; branch employees who conduct data analysis and generate fraud reports; police investigators who examine flagged suspicious cases; and general users with read-only access privileges.The fraud detection model utilizes eight key dataset parameters: DCOD_CRD (Date Code Credit), DCRD_COPD (Date Credit Copy), DPE_COD (Department Code), CDS (Claim Decision Status), PCD (Policy Code), CR (Claim Ratio), PP (Premium Payment), and CCC (Claim Cost Category) to enable comprehensive claim evaluation and risk assessment.Performance evaluation demonstrates that the system achieves 92% accuracy with corresponding precision levels, though recall performance measured at 8% indicates room for improvement in identifying all fraudulent cases. The results confirm successful identification of fraudulent claims while establishing enhanced collaboration frameworks between insurance branches and law enforcement agencies, thereby streamlining investigation processes and improving overall fraud detection efficiency Key Words: Insurance fraud detection, machine learning, Naive Bayes classification, ASP.NET, web-based system, claim analysis etc.