An efficient approach for building customer profiles from business data

Abstract

Data mining (DM) is a new emerging discipline that aims at extracting knowledge from data using several techniques. DM proved to be useful in business where transactional data turned out to be a mine of information about customer purchase habits. Therefore developing customer models (called also profiles in the literature) is an important step for targeted marketing. In this paper, we develop an approach for customer profiling composed of three steps. In the first step, we cluster data with an FCM-based algorithm in order to extract “natural” groups of customers. An important feature of our algorithm is that it provides a reliable estimate of the real number of distinct clusters in the data set using the partition entropy as a validity measure. In the second step, we reduce the number of attributes for each computed group of customers by selecting only the “most important” ones for that group. We use the information entropy to quantify the importance of an attribute. Consequently, and a result of this second step, we obtain a set of groups each described by a distinct set of attributes (or characteristics). In the third and final step of our model, we build a set of customer profiles each modeled by a backpropagation neural network and trained with the data in the corresponding group of customers. Experimental results on synthetic and large real-world data sets reveal a very satisfactory performance of our approach.