Application of fuzzyARTMAP for churn prediction in bank credit cards

Here, we apply fuzzyARTMAP together with feature selection to predict customer churn in bank credit cards. The dataset analysed are taken from Business Intelligence Cup 2004. Since, it is a highly unbalanced dataset with 93% loyal and 7% churned customers, we employed (1) under-sampling, (2) over-sampling, (3) combination of under-sampling and over-sampling and (4) SMOTE for balancing the dataset. We performed ten-fold cross validation throughout. Further, we designed 'union method of feature selection' by considering the union of feature subsets selected by t-statistic and mutual information. Since identifying churner is paramount from business perspective, management accords higher priority on sensitivity alone. Therefore, by considering sensitivity alone, we observed that fuzzyARTMAP performed exceedingly well when preceded by union method of feature selection rather than without it. Further, the proposed method outperformed all techniques employed by Kumar and Ravi when analysing the unbalanced data. This is a significant result of the study.