Data Mining Technique for Predicting Telecommunications Industry Customer Churn Using both Descriptive and Predictive Algorithms

As markets have become increasingly saturated, companies have acknowledged that their business strategies need to focuson identifying those customers who are most likely to churn. It is becoming common knowledge in business, that retainingexisting customers is the best core marketing strategy to survive in industry. In this research, both descriptive and predictivedata mining techniques were used to determine the calling behaviour of subscribers and to recognise subscribers with highprobability of churn in a telecommunications company subscriber database. First a data model for the input data variablesobtained from the subscriber database was developed. Then Simple K-Means and Expected Maximization (EM) clusteringalgorithms were used for the clustering stage, while Decision Stump, M5P and RepTree Decision Tree algorithms were usedfor the classification stage. The best algorithms in both the clustering and classification stages were used for the predictionprocess where customers that were likely to churn were identified.Keywords: customer churn; prediction; clustering; classificatio