MDD-based Domain Adaptation Algorithm for Improving the Applicability of the Artificial Neural Network in Vehicle Insurance Claim Fraud Detection

Insurance fraud detection is a critical task for insurance companies, as fraudulent claims result in financial losses and increased premiums for honest policyholders. Traditional fraud detection methods rely on rule-based approaches and manual investigation, which are limited in their ability to adapt to evolving fraud patterns. In this study, we propose a novel approach using an artificial neural network (ANN) combined with Margin Disparity Discrepancy (MDD)-based domain adaptation to improve the generalization ability of fraud detection models across different datasets. We first preprocess the data by applying K-Means clustering to segment source and target domains based on distribution differences. We then compare multiple machine learning models, including decision trees, random forests, k-nearest neighbors, and gradient-boosted decision trees, finding that ANN achieves the best performance. To further enhance generalizability, we introduce MDD-based domain adaptation, aligning feature distributions between the source and target domains. Experimental results demonstrate that the adapted ANN significantly improves fraud detection accuracy, achieving a higher F1-score and recall while reducing the false negative rate. These findings highlight the effectiveness of domain adaptation in addressing distributional shifts in fraud detection, making the proposed model a promising solution for real-world insurance fraud detection systems.