| Summary: | 博士 === 國立東華大學 === 企業管理學系 === 100 === In previous study, most of existing researches about donor defection have focused on why the donors defect or how to keep them loyal, but what remain explored is that how to predict the likely defecting donors in advance. Hence, in order to help fill this gap on the donor defection, this study utilized a two stages process of dealing with 18 variables within three topics: customer loyalty, socio-demographic information and the duration of donor relationship when data is obtained from a database instead of a survey. At the exploration stage, this study used binary logistic regression to find out the predictive variables; at the prediction stage, we compared several famous classifier techniques including binary logistic regression, back-propagation neural network, support vector machine, decision tree and naïve bayes to construct a donor defection decision support system. In addition, this study designed a misclassification cost measurement by taking type I errors, type II errors and economic cost into account, which is more suitable to evaluate the donor defection prediction model. Finally, results show that decision tree has best performance with three predictors: weighted recency, regular donation and duration of donor relationship.
|
|---|
