A Machine Learning and Explainable Artificial Intelligence approach for Insurance Fraud Classification

Abstract

This study addresses the critical issue of fraud in the vehicle insurance market by introducing a comprehensive framework that employs advanced detection models, including explainable artificial intelligence (XAI) and heterogeneous classifiers to effectively identify fraudulent activities. The incorporation of XAI is particularly noteworthy, as it provides a set of frameworks and resources that enhance the interpretability and transparency of machine learning algorithm detections, a crucial advancement for the integrity and trustworthiness of insurance operations. Our methodology demonstrates the application of three distinct XAI techniques—Shapley Additive Values (SHAP), Explain Like I’m 5 (ELI5), QLattice, and Local Inter pretable Model-agnostic Explanations (LIME)—to eluci date the machine learning model’s decision-making process. This approach not only ensures that our detection models are interpretable but also allow us to identify the most significant factors influencing fraud detection: the age of the vehicle, the base policy, fault, deductible, and the policy holder’s age. The standout contribution of our research is the development and validation of a multi-stack machine-learning model configuration that achieves an unprecedented accuracy rate of 96%, markedly surpassing the performance of conventional classifiers also convey our model interpretation through XAI. This achievement highlights the efficacy and the potential of our proposed framework to revolutionize fraud detection practices in the vehicle insurance sector. By providing a robust, accurate, and interpretable solution to combat fraud, this study makes a significant contribution to the field, offering valuable insights and tools for insurance providers seeking to enhance their fraud detection capabilities. The integration of machine learning and XAI in our framework not only addresses a pressing challenge within the insurance industry but also sets a new benchmark for the development of advanced, reliable fraud detection systems.