A support vector machine (SVM) approach to imbalanced datasets of customer responses: comparison with other customer response models

Abstract

Customer response is a crucial aspect of service business. The ability to accurately predict which customer profiles are productive has proven invaluable in customer relationship management. An area that has received little attention in the literature on direct marketing is the class imbalance problem (the very low response rate). We propose a customer response predictive model approach combining recency, frequency, and monetary variables and support vector machine analysis. We have identified three sets of direct marketing data with a different degree of class imbalance (little, moderate, high) and used random undersampling method to reduce the degree of the imbalance problem. We report the empirical results in terms of gain values and prediction accuracy and the impact of random undersampling on customer response model performance. We also discuss these empirical results with the findings of previous studies and the implications for industry practice and future research.