Finance Department


Date of this Version


Document Type



Journal of Actuarial Practice 11 (2004), pp. 79-102


Copyright 2004 Absalom Press


Statistical methods such as regression and survival analysis have traditionally been used to investigate the factors affecting the duration of terminated life insurance policies. This study explores a different approach: it uses a more recently developed data mining technique called decision trees. By sequentially partitioning the data to maximize differences in the dependent variable (duration in this study), the decision trees technique is good at identifying data segments with significant differences in the dependent variable. This identification can be useful when a company is trying to understand the factors driving or associated with the termination of life insurance policies. Decision trees also have an advantage over other techniques such as linear regression in their ability to detect nonlinear and other complex relationships that are more likely to exist in any practical data set.