Analytical CRM using Soft Computing Techniques: A Study of Implications of Feature Selection

Abstract

Analytical CRM has become the need of the hour as the data generated about customers is huge and imbalanced in nature. However, it becomes nearly impossible to analyse available data manually. Hence, soft computing community has reported various approaches to analyse and understand the data. In this research study we propose the use of Feature Selection approach prior to classification modelling in order to reduce the complexity of the classifier without compromising its performance. We have used Churn prediction in bank credit card customer data which is medium in size and highly imbalance in nature. Bank needs to understand about their customers (valuable) and device policies to retain them who are about to switch their loyalties to the competitor. As the problem at hand is about predicting possible churners, sensitivity is accorded priority in analysing the performance. Based on sensitivity yielded it is observed that the proposed approach i.e., using reduced features yielded better results compared to the results obtained using full feature data. Further, it is also observed that Support Vector Machine (SVM) performed better compared to Decision Tree (DT).