A Case Study of Using Data Mining Techniques in Service Quality-Use Pxmart as An Example

• Main Authors: Yi-Jia Huang, 黃一家
• Other Authors: Hsin-Hung Wu
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/62528344806563921134

碩士 === 國立彰化師範大學 === 企業管理學系 === 100 === This study used the service quality scale for Pxmart established by Chen (2011) and used classification and regression tree (CART) of IBM SPSS Modeler 14.2 to classify and predict the customers’ behaviors. Twenty eight items identified by Chen (2011) are input variables, while the three categories of amount spent per visit is the target variable. This study intends to predict the amount spent per visit and analyze the rules between input and target variables. Later, dimension reduction was performed by SPSS 18.0 to identify the variables with eigenvalue greater than one, and these variables become the input variables for CART. Finally, Bayesian network (BN) was applied to repeat the prior processes, and an evaluation among CART and BN models was performed. In the original CART model, the tree depth was 13, and 33 rules of categories of amount spent per visit were generated. After merging categories of amount spent per visit, the tree depth was 12, 32 rules were obtained, and both results of models are too complex. The tree depth and rules are much simpler after dimension reduction and feature selection with the respective tree depths and rules are 9, 1, 14, and 2. For Bayesian network, the Markov blanket structure in all models could not be analyzed, only TAN (Tree Augmented Naïve Bayes) structure in dimension reduction and feature selection models generate Bayesian network graph. Later, we computed the probabilities of categories of amount spent per visit while consumer satisfaction is given. In model evaluation, the rankings performed by IBM SPSS Modeler 14.2 are: CART model with merged target variable categories, BN model with dimension reduction, original CART model, BN model with feature selection, CART model with dimension reduction, and CART model with feature selection. Finally, we suggested managerial implications for each model.