Forecasting customer's loyalty by means of an unsupervised fuzzy learning method

The use of unsupervised fuzzy learning classifications techniques allows defining innovative classifications to be applied to forecast customers' behavior in retail marketing. Actually, segmenting the clients' portfolio is important for decisionmaking in marketing because it allows discovering hidden profiles which would not be detected with other methods. Different strategies can be established for each defined segment. The objective of this paper is to show the utility of the unsupervised learning techniques applied in marketing segmentation. In particular, the LAMDA algorithm (Learning Algorithm for Multivariate Data Analysis) is used. LAMDA is a hybrid connective-based classification method that combines some of the most interesting capabilities of both purely numeric and purely symbolic algorithms. In order to do ISF 2008 PROGRAM so, it employs the interpolation capabilities of logic operators over fuzzy environments. The algorithm and a set of criteria that permit dynamically select a fuzzy segmentation are being analyzed and implemented over a new Java-based version of the LAMDA algorithm. In this work the concept of adequacy is considered by associating to each one of the patterns a vector of values between 0 and 1. Each one of its components can be considered a degree of membership to each one of the considered segments (classes). The adequacy will describe how well a customer fits in a given segment. The obtained results permit forecasting the degree of adequacy of each customer to each one of the segments given. The unsupervised capability of the LAMDA algorithm along with the analysis tools developed over the system are being applied in the field of marketing segmentation in order to forecast the customer's loyalty of a supermarket chain in Spanish market.