Rough Sets for Database Marketing
Dirk Van den Poel · Studies in fuzziness and soft computing · 1998
This chapter describes how rough sets can be used for response modeling in database marketing. We use real-world data from one of the largest European mail-order companies. Past transaction data of customers, personal characteristics and their response behavior are used to determine whether these clients are good mailing prospects during the next period. We provide a comparison of statistical techniques, machine learning, mathematical programming, rough sets and neural networks in a classification task, and show that rough sets can also be successfully used for response modeling in database marketing. The performance of alternative techniques is judged on the percentage of correct classifications in the validation sample, and on gains chart analysis. The results indicate that on a dataset with only categorical information, the predictive performance of statistical techniques, machine learning techniques and neural networks on a validation dataset is very similar Still the observed differences are significant.