The Research On Auto Insurance Fraud Identification Based On GA-PSO-BP Neural Network

Nowadays,with the rapid improvement of China's economic strength,the number of motor vehicles is also increasing,and auto insurance steadily occupies the position of the largest insurance category in property insurance.However,in the insurance market of our country,the phenomenon of fraud occurs from time to time.According to statistics,the amount involved in fraud cases accounts for at least 20% of the total amount of claims,resulting in losses exceeding 20 billion yuan each year.It can be seen that how to accurately and effectively identify auto insurance fraud is a problem that every property insurance company needs to solve.This dissertation defines the concepts of auto insurance fraud and expounds the asymmetric information theory?incomplete contracting theory and game theory.Then this thesis analyzes causes and harms from the perspectives of the insured,the insurer and the social environment,and summarizes the characteristics of auto insurance fraud from three aspects: diversification of fraud forms,specialization of fraudulent means and the organization of criminal subjects.Finally,it introduces and analyzes auto insurance fraud detection model.On this basis,this dissertation uses BP neural network model to carry out auto insurance fraud detection analysis using a public dataset simulated by descriptive and statistical features of the Massachusetts AIB claims dataset as sample data.The empirical results show that the overall prediction accuracy rate using the BP neural network model is 87.5%,especially the prediction accuracy rate of fraud claims is low,only 82%,which can not identify the fraud claim cases in the sample data more accurately.Based on this problem,this thesis designs and implements GA-PSO-BP neural network model based on a combination of particle swarm optimization and genetic algorithm.The model optimizes the initial connection weights and thresholds,solves the problem of converging to local extremes and improves the performance of the neural network.The empirical results show that the overall prediction accuracy rate of the optimization model increases to 94.5%,especially for fraud claims,which increases significantly to 95%.It indicates that the combination of particle swarm optimization and genetic algorithm can optimize the BP neural network model effectively.