Churn Prediction Analysis by Combining Machine Learning Algorithms and Best Features Exploration

Abstract

The market competition and the high cost of acquiring new customers have led financial organizations to focus more and more on effective customer retention strategies. Although the banking and financial sectors have low churn rates compared to other sectors, the impact on profitability related to losing a customer is comparatively high. Thereby, customer turnover management and analysis play an essential part for financial organizations in order to improve their long-term profitability. Recently, it appears that using machine learning to predict churning improves customer retention strategies. In this work, we discuss some specific machine learning models proposed in the literature that deal with this problem and compare them with some emerging models, based on Ensemble learning algorithms. As a result, we build a predictive churn approaches that look at the customer history data, check to see who is active after a certain time and then create models that identify stages where a customer can leave the concerned company service. Ensemble learning algorithms are also used to find relevant features in order to reduce their number which is of great importance when performing the training step with some classical models such us Multi-Layer Perception Neural networks. The proposed approaches can achieve up to 89% in accuracy when other research works, dealing with the same dataset, can achieve less than 86%.