Automatic Discovery of Potential Causal Structures in Marketing Databases Based on Fuzzy Association Rules

Marketing-oriented firms are especially concerned with modeling consumer behavior in order to improve their information and aid their decision processes on markets. For this purpose, marketing experts use complex models and apply statistical methodologies to infer conclusions from data. In the recent years, the application of machine learning has been identified as a promising approach to complement these classical techniques of analysis. In this chapter, we review some of the first approaches that undertake this idea. More specifically, we review the application of Fuzzy-CSar, a machine learning technique that evolves fuzzy association rules online, to a certain consumption problem analyzed. As a differentiating sign of identity from other methods, Fuzzy-CSar does not assume any aprioristic causality (so model) within the variables forming the consumer database. Instead, the system is responsible for extracting the strongest associations among variables, and so, the structure of the problem. Fuzzy-CSar is applied to the real-world marketing problem of modeling web consumers, with the aim of identifying interesting relationships among the variables of the model. In addition, the system is compared with a supervised learning technique, which is able to extract associations between a set of input variables and a pre-fixed output variable, expressly designed for this marketing problem. The results show that Fuzzy-CSar can provide interesting inforAlbert Orriols-Puig Grup de Recerca en Sistemes Intel·ligents, Enginyeria i Arquitectura La Salle (URL), 08022 Barcelona (Spain). e-mail: [email protected] Jorge Casillas Department of Computer Science and Artificial Intelligence, Universidad de Granada, 18071 Granada (Spain). e-mail: [email protected] Francisco J. Martı́nez-López Department of Marketing, Universidad de Granada, 18071 Granada (Spain) and Universitat Oberta de Catalunya, 08035 Barcelona (Spain). e-mail: [email protected]