The Study of Prediction Model for Life Insurance Premium Income Based on Machine Learning Algorithms

• Main Authors: Chen, WEI-HSIU, 陳威秀
• Other Authors: Lin, Wen-shiu
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/dptvwf

碩士 === 輔仁大學 === 資訊管理學系碩士班 === 105 === In recent years, the insurance market in Taiwan has become increasingly competitive because of the economic development, the growth of risk awareness, the relaxation of law and the trend of banking mergers. According to competitive pressure of the insurance market, the accurate prediction of national premium income has become the key success factors of business strategy for the insurance company. Aadditionally, the past studies has point out that the premium income is related to macroecnomic phenomenon. As a result, we use macroeconomic Index, basing on insurance forecasting technology and machine learning algorithms to build a forecasting model in premiums income of personal insurance. The algorithms which been used are as follows：Logistic Regression, K-Nearest Neighbor, Back Propagation Neural Network, Support Vector Regression, Classification And Regression Trees, and Random Forest. The result we found are listed below: 1. For the time period of forecasting, when the time period is two years, forecast result is better than others. However, different algorithms have their appropriate time period. 2. For validity, the average mean squared error of Random Forest, K-Nearest Neighbor , Logistic Regression and Classification And Regression Trees are better than others. 3. For reliability, Logistic Regression, Support Vector Regression and K-Nearest Neighbor are more stable than others. 4. For efficiency, Logistic Regression, Classification And Regression Trees, Support Vector Regression and K-Nearest Neighbor are more efficient than others. They are suitable for applying to real-time online system. 5. According to all the above factors, the author thinks K-Nearest Neighbor and Classification And Regression Trees are suitable for building a forecasting model in premiums income because of the good perfomace in the validity, reliability and efficiency.