Hybrid C&RT-Logit Models In Churn Analysis

This article attempts to explain and predict the termination of relationships in telecommunications services by using the hybrid C&RT-logit model. The combination of decision trees (C&RT algorithm) with the logistic model enriches the model interpretation and sometimes improves the accuracy of prediction. Decision trees permit to detect interactions among variables and make the model resistant to outliers and to lack of data. On the other hand, the logistic model can extend the interpretation by using odds ratios. The solution delivered by the hybrid approach was compared with the decision tree model and the logistic model. Due to the difficulty in obtaining the real dataset from the Polish market, it was decided to build a model based on the data obtained from the repository http://www.dataminingconsultant.com/DMMM.htm . The models’ performance was estimated by using popular measures such as accuracy, recall, precision, true negative rate, G-mean, F measure and lift charts.