Insurance claim fraud detection

"Insurance Claim Fraud Detection" is a conclusive proof of concept for high-intelligence analytics incorporation in the financial services domain as a whole. In fact the success of this project in shifting its primary operation mode from the traditional manual heuristic audit approach to an active analytics-driven approach is a viable antidote to the ubiquitous problems of fiscal leakage and stagnation. Project Achievements and Technical Efficacy The architecture thus developed was able to showcase a strong capability of handling high-dimensional real-world datasets. The architecture underwent a comprehensive life cycle of refinement of the data using a Random Forest ensemble to achieve a highly sophisticated trade-off of Sensitivity (Recall) and Predictive Precision. In the evaluation phase it is ensured that the model is not "memorizing" past records and is actually detecting the "behavioral markers" associated with fraudulent activity. Thus it is ensured that the system is an efficient deterrent even as the actions associated with fraudulent records are thwarted thus resulting in the "false alarm" rate that is characteristic of rule-based filters being minimized.

1.                [1] Han J. Kamber M. & Pei J. Data Mining: Concepts and Techniques. Morgan Kaufmann Publishers 3rd Edition. 2.                Géron A. Hands-On Machine Learning with Scikit-Learn Keras and TensorFlow. O'Reilly Media. 3.                Phua C. Lee V. Smith K. & Gayler R. "A Comprehensive Survey of Data Mining-based Fraud Detection Research." arXiv preprint arXiv:1009.6119. 4.                Insurance Information Institute. "Background on: Insurance Fraud." Accessed online at iii.org. 5.                Scikit-Learn Documentation. "User Guide: Supervised learning." Accessed online at scikit-learn.org.