A Detailed Procedure for Using Copulas to Classify E-Business Data

Decision support systems are widely implemented to effectively utilize the tremendous amount of data generated by information systems throughout an organization. In one common implementation, the goal is to correctly classify a customer so that appropriate action can take place. This may take the form of a customized purchase incentive given to increase the probability that a transaction is completed, while enhancing profitability. Intelligent agents employing neural network technology that function as Bayesian classifiers are one approach used here. Another approach that has been around for decades, called copulas, to our knowledge has yet to be utilized for classification in e-business applications. Copulas are functions that can describe the dependence among random variables. The very fact that copulas directly address co-dependence among variables may make them especially attractive in e-business applications where large numbers of correlated attributes may be present that could negatively affect the performance of other methods. In this paper, the basics of Bayesian decision making and posterior probabilities are reviewed. A detailed procedure for using copulas as Bayesian classifiers for e-business data is presented. The emphasis in describing the method is placed upon practitioner understanding to facilitate replication in real situations while maintaining technical rigor to ease computerized implementation