Predictive Profiles for Transaction Data using Finite Mixture Models

Massive transaction data sets are routinely recorded in a variety of applications including telecommunications, retail commerce, and Web site management. In this paper we address the problem of inferring models from such transaction data in the form of predictive profiles of individual behavior. We describe a generative mixture model that accounts for population heterogeneity in transaction generation. An approximate Bayesian framework is used for parameter estimation that combines an individual’s specific history with more general population patterns. The proposed model is shown to consistently outperform non-mixture and non-Bayesian techniques in predicting out-of-sample individual behavior on two large real-world transaction data sets.