The Comparison Between the Prediction Power of Neural Networks and Logistic Regression Analysis on the Auto Physical Insurance Claim

• Main Authors: Wei-Ling Chang, 張緯翎
• Other Authors: none
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/36990242451326370737

碩士 === 逢甲大學 === 保險所 === 93 === This study uses the private auto physical insurance policy holders of a property and casualty insurance company in Taiwan as the research objects, and divides the data into sets of in-the-sample and out-of-the-sample data. We, first, set up the claim prediction model by in-the-sample data, and then test the prediction power by out-of-the-sample data. Finally, the prediction power for the private auto physical insurance claim by logistic regression and neural networks analysis is carried out by confusion matrix. Our empirical results show that the determinants for private auto physical insurance claim of logistic regression and neural networks analysis are quite consistent. Besides the current premium factors for private auto physical insurance, including the variables of insurance type, exhaust volume, insurance beginning year, and living area can increase the prediction correct percentage for insurance claim. The overall prediction power of neural network analysis is better than logistic regression analysis for in-the-sample data, but worse for out-of-the sample data.