A major challenge in direct marketing is to build a customer-selection model that can help achieve higher response rate and greater profit at the same time. In this study, I adopt a bi-objective optimization (BOO) approach and propose a two-stage method using support vector machine (SVM) and support vector regression (SVR) to maximize response rate and profit simultaneously. To deal with the difficulty of learning models from imbalanced data, synthetic minority over-sampling technique (SMOTE) is used to generate more balanced datasets. Experiments are conducted on two datasets, a direct marketing dataset and the KDD-98 dataset, to compare the predictive performance of the two-stage BOOSVM with other benchmark methods including logistic regression and the parallel Multi-objective Evolutionary Algorithm (MOEA). The results of decile analysis suggest that the proposed two-stage BOOSVM model with SMOTE method is more effective and efficient than the competing models in improving response rate and profitability.
|Date of Award||2013|
- Department of Marketing and International Business
|Supervisor||Geng CUI (Supervisor) & Man Leung WONG (Supervisor)|