Framework for Analysis and Detection of Fraud in Health Insurance
Nirmal Rayan · 2019
The health insurance industry generates a wide range of data from patients' information to provider payment and claims report. The impact of fraud, waste, and abuse (FWA) in medical management is on the rise and contributes significantly to the increase in cost. Traditional methods of handling fraud include human inspection and heuristic rules. They are time-consuming, impractical and insufficient. Data Mining and Machine Learning play a dominant role in detecting and preventing fraud. We explore the use of statistical methods to create a rule-based heuristic engine that works with self-learning Decision Trees. This paper introduces a hybrid framework that combines domain expertise (Rule Engine), supervised learning (Decision Trees & Averaged Perceptron) and unsupervised learning (outlier analysis, k-means Clustering) techniques to identify fraudulent claims from a given set of outstanding claims. The investigation team is intimated with a weighted priority queue of outstanding claims listing the most-likely fraudulent claims with remarks for proactive and retrospective analysis. Our initial case study with one insurer demonstrates an increase in hit-rate by 209.4%.