순차적 자료융합방법을 이용한 은행고객의 가치예측

초록

This study applies data fusion approach to predict the value of bank customers. The three approaches applied include the Euclidean distance-based fusion, the Mahalanobis distance-based fusion, and the regression/logit-based fusion. The data consist of 1,000 customers with their activities associated with various accounts, demographic variables, and class segmentation for the fiscal year of 2006 and 2007. For the comparison of model performance, MAD and the paired samples t-test are used for the 10 numeric variables, while correct classification ratio and the McNemar test are used for the three categorical variables. The experiment results show that with respect to MAD, model performance is superior in the order of the Euclidean model, the Mahalanobis model, and the regression model. Correct classification ratio is best for the logit model, and the Euclidean model and the Mahalanobis model follow the next. The contribution of the current study is that we attempt to extend the scope of data fusion into sequential data fusion, which is necessary for the systematic data accumulation and analysis under CRM strategy.

키워드