Enhancing Fraud Detection with Advanced Anomaly Detection for a Leading P&C Carrier
Overview.
A leading US-based property and casualty (P&C) insurance carrier sought to improve fraud detection and reduce false positives in claims processing. Coforge implemented advanced anomaly detection models and external data integration, resulting in significant improvements in recall, precision, and faster claim processing. The solution also included an interactive dashboard for investigators to evaluate high-risk agents and policies.
The client faced significant challenges in detecting fraudulent claims:
Fraud Detection: Needed to detect fraudulent claims more accurately and reduce false positives generated by existing heuristic methods.
Straight Through Processing: Sought to establish a straight-through process for select categories of claim transactions.
Our Solution.
Coforge implemented a comprehensive anomaly detection solution:
Class Imbalance Treatment: Deployed sophisticated techniques such as SMOTE, Tomek Links, and NearMiss to treat class imbalances.
Feature Engineering: Introduced external data sources to enhance feature engineering.
Anomaly Detection Models: Developed unsupervised anomaly detection models using techniques like One Class SVM, Local Outlier Factor, Isolation Forest, and K Nearest Neighbour.Combined these with a rule-based statistical model to classify claims into Low, Medium, and High fraud risk buckets.
Interactive Dashboard: Helped the client allocate resources more efficiently among various customer segments.
Key Highlights:
Coforge's solution delivered significant value to the client's operations:
Improved Recall and Precision: Achieved approximately 25% improvement in recall and 12% lift in precision over the previous model.
Faster Claim Processing: Reduced claim processing time by approximately 15%.
Risk Classification: Enabled classification of claims into different fraud risk buckets.
Enhanced Investigator Tools: Provided an interactive dashboard for better evaluation of high-risk claims.
