Document Type : Original Research Paper
Authors
1 Department of Actuarial, Eco Insurance Institute of Higher Education, Allameh Tabatabai University, Tehran, Iran
2 Department of Actuarial, Social Security Organization, Tehran, Iran
Abstract
Fraud in unemployment insurance rights and benefits is always one of the sensitive and important issues in the field of social insurance, which according to the laws is a criminal offense and can be prosecuted. Currently, the best method to evaluate fraud is to control it in the initial stages of its formation and with the help of the information of discovered frauds in the past. In this article, first, the standard steps of fraud control in insurance claims are examined, and then, considering the existence of a suitable database regarding unemployment insurance recipients of the Social Security Organization, two data mining methods, neural networks and decision trees, are used in order to find suitable patterns. It has been shown that it can be a useful tool with a significant reduction in time and cost to help in the timely evaluation of fraud in unemployment insurance claims. In the process of the experimental study, these methods have been tested on real data including information on 15,983 new and current unemployment insurance claims and the efficiency of each method has been measured. The neural network method with an accuracy of 88% in assessing whether the demands are correct or normal has achieved the best performance in comparison with the decision tree method with an accuracy of 84%. Based on this, the most important variables affecting fraudulent claims in the neural network method are, respectively, the variables of the insured's previous job, insurance premium payment history, age, and in the decision tree method, the variables of the geographical location of the branch, gender, and number dependents of the applicant.
Keywords
)
Letters to Editor
Send comment about this article