Secure Score: AI-Based Insurance Policy Analytics System

Author

Ms. Sini Prabakaran, Kanishkaa.S, Karpagalakshmi A, Mathumitha B, Mithra Devi M


Abstract

 Insurance companies process a large number of claims and policy documents, making claim verification and risk assessment important parts of insurance management. Manual verification can be time-consuming, particularly when reviewers need to examine supporting documents, identify repeated claims, and investigate unusual billing patterns. To address these challenges, this project presents SecureScore, an AI-based insurance policy analytics system designed to support insurance claim analysis through machine learning and document processing.

The proposed system combines claim classification, duplicate claim detection, billing anomaly screening, document information extraction, and policy information retrieval within a unified platform. A Random Forest classification model is used to analyse claim data, while a Flask backend connects the analytical components with a React and TypeScript user interface. The system also includes document chunking, vectorization, and evidence retrieval to help users locate relevant information in policy documents.

The classification model was evaluated using a synthetic health insurance claims dataset containing approximately 19,962 records and 37 features. The experimental results achieved a ROC-AUC score of 0.885, accuracy of 0.869, precision of 0.762, recall of 0.688, and F1-score of 0.724. These results demonstrate the model's classification performance on the experimental dataset. Further evaluation using suitable real-world data is required before applying the findings to actual insurance claims. SecureScore aims to support human reviewers by organizing claim information and identifying records that may require further investigation.


Keywords

SecureScore, Insurance Analytics, Machine Learning, Random Forest, Claim Verification, Duplicate Detection, Anomaly Detection, Policy Information Retrieval.



Full Text:

Download Paper PDF


References


  1. Breiman, L. “Random Forests.” Machine Learning, vol. 45, pp. 5–32, 2001.
  2. Saito, T., and Rehmsmeier, M. “The Precision-Recall Plot Is More Informative than the ROC Plot When Evaluating Binary Classifiers on Imbalanced Datasets.” PLOS ONE, vol. 10, no. 3, 2015.
  3. Lundberg, S. M., and Lee, S.-I. “A Unified Approach to Interpreting Model Predictions.” Advances in Neural Information Processing Systems, vol. 30, 2017.
  4. Pedregosa, F., et al. “Scikit-learn: Machine Learning in Python.” Journal of Machine Learning Research, vol. 12, pp. 2825–2830, 2011.
  5. Flask Project. Flask Documentation.
  6. React Team. React Documentation.

 

 

 

Share your valuable work from Social Media Buttons