FraudNet: A Network-centric Method for Balancing and Detecting Anomalies in Healthcare Insurance

Healthcare insurance fraud detection is challenged by the prevalence of class imbalance in transaction datasets, which undermines the accuracy of predictive models. Traditional data balancing methods have been insufficient in addressing the complexity of healthcare fraud. This study proposes a novel framework, FraudNET, which utilizes a hybrid of Generative Adversarial Networks (GANs) and Convolutional AutoEncoders (CAE) to generate a balanced dataset and enhance the detection of healthcare insurance fraud. The performance of FraudNET was evaluated against traditional models using metrics such as Precision, Recall, F1 score. FraudNET significantly improved the balance of the dataset, with the GAN generated synthetic data exhibiting high fidelity. The CAE model demonstrated superior performance in classifying fraudulent transactions, outperforming established methods like Synthetic Minority Over-sampling Technique (SMOTE) and Adaptive Synthetic (ADASYN) in comparative analyses. The integration of GANs and CAE in the FraudNET framework offers a promising solution to the challenge of class imbalance in healthcare insurance fraud detection.