Improvement of customer response prediction in direct marketing by neural networks

Abstract

The purpose of this study is to identify the potential customers to address in direct marketing programs, which has been regarded as one of the most important issues for direct marketers. The most important matter is the customer’s data set, which is always highly imbalanced. In this study, by combining the random under-sampling and over-sampling of the majority and minority classes that have been used frequently in past studies, we have designed and developed a dynamic and effective algorithm to identify and predict potential customers by clustering customers and extracting more balanced samples. For this purpose, a travel agency database (of over 10,000 records) has been used. The results indicate that customer's raw data cannot make a reliable prediction. On the other hand, Re-sampling methods using customer clustering and combining of minority and majority classes according to the proposed algorithm dramatically increases the prediction power of the decision tree and can be used in different situations and markets. Finally, by combining the results of the extracted XML codes and "multiple" criterion in each step, we can identify and rank potential customers and target them in an efficient way.

Keywords

• prediction optimization
• class imbalance
• data mining
• neural network