Affiliations: [a] College of Mathematics and System Science, Shandong University of Science and Technology, Qingdao, Shandong 266590, China | [b] College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao, Shandong 266590, China
Correspondence:
[*]
Corresponding author: Wei Liu, College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao, Shandong 266590, China. E-mail: [email protected].
Abstract: With the rapid development of China’s insurance industry, insurance fraud incidents are also increasing, especially in the field of auto insurance. Therefore, the vehicle insurance fraud identification model based on extreme learning machine is studied. Because the initial connection weight and hidden layer neuron threshold of the ELM are generated randomly, the recognition results are unstable and the accuracy is affected. Therefore, artificial fish swarm algorithm is used to optimize the model parameters. This paper adaptively improves the step size, visual field and crowding degree of artificial fish swarm. First of all, the principal component analysis method is used to generate the input vector of the ELM model for vehicle insurance fraud. Then the weights and thresholds of the ELM model are optimized by improved artificial fish swarm algorithm. Finally, the model is applied to vehicle insurance fraud identification. The empirical analysis shows that the optimized model has less recognition error and higher recognition stability compared with the traditional ELM classification model.
