Social Network Analytics for Supervised Fraud Detection in Insurance.

Area of Science:

• Data Science
• Network Analysis
• Insurance Fraud Detection

Background:

• Insurance fraud involves exaggerated or intentionally damaged claims.
• Traditional fraud detection methods often rely solely on claim-specific data.
• Understanding the social connections surrounding insurance claims can reveal fraudulent patterns.

Purpose of the Study:

• To develop a fraud detection strategy leveraging social network analysis.
• To assess the effectiveness of network-derived features compared to traditional features.
• To improve the accuracy and efficiency of identifying fraudulent insurance claims.

Main Methods:

• Constructing a social network linking claims and involved parties (policyholders, brokers, experts, garages).
• Applying the BiRank algorithm with a fraud-specific query vector to calculate fraud scores.
• Extracting network-based features (fraud scores, neighborhood structure) and combining them with claim-specific features for a supervised learning model.

Main Results:

• Models incorporating network features demonstrate strong performance in fraud detection.
• Network-derived features outperform traditional claim-specific features alone.
• Combining network and claim-specific features further enhances the performance of supervised fraud detection models.

Conclusions:

• Social network analysis is a valuable tool for enhancing insurance fraud detection.
• The proposed method effectively flags suspicious claims for further investigation.
• This approach contributes to a more guided and intelligent fraud investigation process.