Using Data Mining Techniques for Classifying the Potential Value of Policyholders

• Main Authors: CHOU-YIN YANG, 楊卓穎
• Other Authors: HONG-TU LIAO
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/61685829799451070788

碩士 === 世新大學 === 資訊管理學研究所(含碩專班) === 97 === After joining WTO and the government passed the bill for financial holding companies, the life insurance industry of Taiwan falls into keen competition. Under the circumstances, how should one respond to fast changing insurance policies, satisfy customers’ massive demands to protect their families, promote the company’s competitiveness in the market, and explore the potential value of life insurance? All of these have become extremely important topics at present. Other than seeking for new customers and decision-making for realization of new policies, continuing to satisfy the life insurance company’s original policy-holders and obtain greatest benefits for both company and customers is an important task to be worked on. Then one may break through multiple thoroughfare pinches, release from the competition pressure of weeding out the old items and bringing forth the new ones, and implement the effective marketing tactics. The major purpose of this thesis is to mine for the best potential value of existing policy-holders and try to find the possible best classification of them. The result of this study will provide the life insurance company a good reference in merchandising new products. The source of research is based on a large database of policy-holders from a particular life insurance company in Taiwan. The data we used is from January 1, 2006 to September 30, 2008. The data of 2006, 2007 and that of 2006 plus 2007 are used as training data models (Learning Model), and then the annual data of individual years 2007 and 2008 are used as testing data models (Testing Model). The major research technology is data mining, and research method is implementation of three algorithms on the Weka platform. The algorithms we used are J48 (Decision Tree), Naïve Bayes and K-menas. By comparing the data from learning model and testing experimental model, we may understand which algorithm can predict the best classification of insurance policies. This classification implies the best potential value of existing policy-holders, and provides a good reference for the insurance company in decision-making for realization of new policies.