FCP Filter: A Dynamic Clustering-Prediction Framework for Customer Behavior

Abstract

Customer purchase behavior prediction plays an important role in modern retailing, but the performance of this task is often limited by the randomness of individual historic transaction data. In the meanwhile, Fragmentation and Coagulation Process (FCP), a stochastic partition model, has recently been proposed for identifying dynamic customer groups and modeling their purchase behavior. However, FCP is not able to forecast the purchase behavior because such a data-driven method requires transaction observations to conduct clustering. To tackle this challenge, we propose FCP filter, a clustering-prediction framework based on FCP, which can forecast purchase behavior and filter random noise of..