Detection Of Insurance Fraud By Partial Sparse SVM

ABSTRACT:Since the reform and opening up, China’s rapid development of the insurance industry had made remarkable achievements. Due to the lack of legal regulations and supervision methods, the incidence of insurance fraud has risen considerably. The problem of insurance fraud is also an international problem. It’s essential for domestic insurance companies to establish a scientific insurance fraud detection model which can identify insurance fraud effectively. Using support vector machine theory we can establish a quadratic programming model to solve the recognition problem of insurance fraud. The number of the ill policy in the historical data is significantly less than the number of normal policy, however, it is very difficult to identify the ill policy in the prediction of the model leading to a poor warning result. In this paper, we propose a quadratic programming problem with a partial sparse constraint condition to reduce the number of normal policies in historical data, which makes the data set more balanced and makes the model more effective. Furthermore, we give a sparse projection algorithm, and prove the convergence of the algorithm. In addition, the numerical experiments show that the proposed algorithm has better numerical performance than the existing algorithms.