A Hybrid Model for Detecting Insurance Fraud Using K-Means and Support Vector Machine Algorithms

Abstract

Private stakeholders and governments across the globe are striving to improve the quality and access of healthcare services to citizens. The need to improve healthcare services, coupled with the increase in social awareness and improvement of people’s living standards, has seen an increase in medical policyholders in the insurance industry. Even so, the healthcare sector is grappled with increased costs every other year, leading to revision of premiums and increased costs for the policyholders. One of the main factors contributing to the increased costs is fraudulent claims raised by the service providers and the policyholders, leading to unprecedented risks and losses for insurance firms. The insurance industry has set up fraud detection and mitigation systems to mitigate losses brought about by fraudulent claims, which come in two flavors: rule-based systems and expert claims analysis. With rule-based systems, conditions such as missing details, location of the claim vis a vis the location of the policyholder, among other rules, are evaluated by systems to assess the validity of the claims. On the other hand, insurance firms rely on the human intervention of experts using statistical analyses and artificial rules to detect fraudulent claims. The rule-based and expert analysis methods fail to detect patterns or anomalies in claims, which is central to efficient fraud detection. Data mining and machine learning techniques are being leveraged to detect fraud. This automation presents enormous opportunities for identifying hidden patterns for further analysis by insurance firms. This research aims to analyze a hybrid approach to detect medical insurance fraud using both K-Means (unsupervised) and Support Vector Machines (supervised) machine learning algorithms.