Automatic Detection of Suspected Fraud In Motor Insurance Claim Documents

Insurance fraud is one of the most costly scourges for the insurance sector today. Each year, it generates considerable financial losses that permanently undermine the economic stability of companies. Faced with this persistent challenge, insurance companies still struggle to develop reliable and automated tools capable of distinguishing a legitimate claim from a fraudulent one. This article proposes an approach based on visual and textual consistency analysis to identify inconsistencies between the damage observed in images of damaged vehicles and the information reported at the time of the accident. At the heart of our system is an automatic damage detection model based on RAkEL, a multi-label classification approach particularly well-suited to the complexity of automotive claims. This model communicates with a conformity assessment algorithm, which measures the consistency between what the vehicle images reveal and what the insured party reports. Far from relying solely on automation, the system also incorporates the expert opinion of an insurance professional, ensuring a nuanced analysis of ambiguous cases. Furthermore, a comparative memory mechanism cross-references each new claim with previous claims to detect potential duplicates or attempts at repetitive reporting. Several similarity thresholds ranging from 50% to 90% were tested to measure the system’s performance in terms of accuracy, recall and F1 score. The results obtained show that the proposed approach brings an improvement in the identification of suspicious statements.