An Interest Support Based Subspace Clustering Approach to Predicting Repurchases

Abstract

We propose an interest support based subspace clustering approach to improving the performance of predicting interesting behaviors. The approach uses particular heuristics to discover interesting subspace clusters more efficiently and selects a proper set of subspace clusters to build a cluster based classification model. We evaluate our approach using synthetic data and a real world application of predicting customer repurchase behavior. The experimental results show our approach is effective in discovering interesting and interpretable clusters that provide better prediction on customer repurchase behavior.