Preventing customer churn by using random forests modeling

In this paper, we use the improved balanced random forests(IBRF) to predict the customer churn, while integrating a sampling technique and cost-sensitive learning into the standard random forests to achieve a better performance than most existing algorithms. The nature of IBRF is that the best features are iteratively learned by altering the class distribution and by putting higher penalties on misclassification of the minority class. Applied to a credit debt customer database of an anonymous commercial bank in China, they are proven to significantly improve prediction accuracy comparing with other algorithms, such as artificial neural networks, decision trees, and class-weighted core support vector machines (CWC-SVM). The assessment and comparison of these algorithms are made to analyze the traits of them. Data processing and sampling scheme are also detailed introduced.