Healthcare Predictive Modeling for Identifying Fraud in Medical Insurance Claims

Fraud detection in healthcare insurance claims is of prime importance to financial stability, operational efficiency, and policyholder trust. Rule-based and hand-crafted manual audit checks, which are traditional fraud detection methods, produce low quality false positives and low response rates to emerging trends in fraud schemes. This work proposes an integrated scheme of XAI-based and machine learning-based fraud detection towards improved accuracy, explainability, and real-time fraud detection capability. The article proposes a comparison of machine learning algorithm-based schemes, i.e., Logistic Regression, SVM, Random Forest, KNN, and Autoencoder, on fraudulent healthcare claim detection in artificial National Health Insurance System (NHIS) datasets. Experimentation results indicate that highest precision and accuracy of (1.000 and 88.7%, respectively) are produced by Logistic Regression and SVM, which are highly reliable in minimizing false positives. Based on results presented, it is concluded that an integrated fraud detection scheme, consisting of a supervised and an unsupervised learning scheme, can improve fraud detection accuracy significantly. Except for the healthcare sector, the proposed mechanism can be effectively applied to banks, retailing, e-commerce, telephony, and supply chains, wherever fraud detection capability is of particular concern. In addition to the machine learning algorithm, the article presents prime concerns on data privacy-related issues, model interpretability issues, and associated computational complexities in providing inputs towards future directionality in AI-sustained fraud avoidance.