Abstract: In order to well resolve the highly skewed class distribution between churns and no-churns, the customers churn prediction model is realized according to the CRISP-DM (Cross-industry Standard Process for Data Mining) framework. The multi-classifier class-combiner approach is adopted. The model could not result in a 'null' prediction system that simply predicts all instances as non-churners. Compared with a single classifier, the accuracy and generalization of the model are improved.

Key words: customers churn prediction, decision tree, multi-classifier class-combiner, data mining