AI Applications in Commercial Insurance Claims Management and Fraud Detection

Abstract

Detecting fraudulent claims in the insurance industry is of great significance and remains challenging due to the complex nature of the fraud schemes and the sophisticated way of submitting claims. Online retail insurance, which protects sellers from goods return fraudulent claims, is designed for the business scenario of e-commerce platforms. The personal personal-item share information of claimants is defined to indicate a new network between claimants to describe potential interactions. The claimants are represented by network nodes, and the network feature is the personal-item share information. Network learning that leverages the network information is developed to capture sophisticated information. The above procedures combined with a basic classifier successfully identify regular users and fraudsters, achieving a precision of over 80%.
AI applications, especially deep learning architectures for text embeddings, are presented to improve the speed and quality of fraud detection. The architectures process unstructured information and detect fraud with machine learning and statistical models. As transaction data in the finance industry have become larger and more complex given the rise of digital channels, most fraud detection systems need to evolve for efficiency in terms of speed and increasing false rejection. Unstructured information raises many challenges, and deep learning methods that outperform other machine learning and statistical models for analyzing unstructured data are developed. An end-to-end solution is presented to analyze fraud with enhancements on different levels, from embeddings to quantitative characteristics with specific enhancements for fraud prediction.