Using Genetic Algorithm and Decision Tree in the Second Time Sale of Life Insurance

• Main Authors: Kuo Hsiang Chang, 張國祥
• Other Authors: S. W. Lin
• Format: Others
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/36607152866287491535

碩士 === 長庚大學 === 資訊管理學系 === 98 === ABSTRACT With an increasing number of national insurance entities, the industry has been flooded with a multitude of policies offered by companies, trying to increase their market share. At this point, these policies may be inadequate to meet customer demand, and have been steadily replaced with improvements in data mining techniques. Data mining is now widely used in the financial insurance industry. This thesis used a C4.5 Decision Tree (DT) to find customers who would potentially purchase a second policy, and determine the category of coverage from existing customer data. A genetic algorithm (GA) was then applied to select benificial features and to identify the optimal C4.5 parameters; a process, referred to as GA+DT. This thesis used three methods for testing: the default settings of C4.5, GA+DT without feature selection, and GA+DT with feature selection. The results indicated that the highest accuracy rate for the three methods was GA+DT with feature selection, at 67.879%. Insurance companies should collect detailed features of their customer to aid in data mining. In this manner, GA+DT could be effective in determining the optimal parameters and increasing the accuracy of information classification. Keyword：Insurance Sales, Decision Tree, Genetic Algorithm, Feature Selection