Multiclass Fraud Detection in Medical Insurance Claims Using LightGBM and Multilayer Perceptron

Medical insurance fraud greatly affects health care expenditures which cause the insurance company and policy-holders to pay in excess. Fraud detection approaches that are based on several traditional methods are incapable of preserving varied fraud claim types. In this work, we introduce a multi-stage fraud detection framework which unites LightGBM and Multilayer Perceptron (MLP) of two weak classifiers in dual-stage detection with an intention to increase the accuracy of fraud detection and classification. The first one is using LightGBM-which is an ensemble model with gradient boosting capabilities-to decide if a claim is fraudulent. Fraud type (upcoding, phantom billing or duplicate claims) then goes into the MLP model if fraud is detected. This ensemble method in the solution can help increase the efficiency of detecting complex patterns in big data, keeping computational efficiency a balance. The system proposed here provides a scalable, cost effective and flexible answer to the problem of fraud in medical insurance claims which can help insurance providers and policyholders.