Direct marketing is concerned with which customers are more likely to respond to a product offering or promotion. A response model predicts if a customer is going to respond to a product offering or not. Typically historical purchase data is used to model customer response. In response modeling customers are partitioned in to two groups of respondents and non-respondents. Generally, the distribution of records (respondents and non-respondents) in marketing datasets is not balanced. The most common approach to solve class imbalance problem is by using sampling techniques. However; sampling techniques have their shortcomings. Therefore, in this research we integrated supervised and unsupervised learning techniques and presented a novel approach to address class imbalance problem. Recently hybrid data mining techniques have been proposed and they intend to improve the performance of the basic classifiers. Finally we compared the performance of the hybrid approach to that of the sampling approach. We could show that the hybrid ANN model achieved higher prediction accuracy and higher area under the curve value than the corresponding values of the bagging neural network which was trained based on the sampled training set.