Leveraging the efficiency of Ensembles for Customer Retention

To attract more customers every bank comes up with new offers every day. Due to this a customer is highly likely to get churned if the user gets a better offer at another bank. To survive in this competition, banks need to be updated regarding the offers present in market as well as how much their customers are loyal to their services. Customer demographics and credit card usage details are significant parameters to analyze customer behavior in the banking sector. The selected dataset aligns with these parameters but is highly unbalanced, which may produce skewed results. To tackle this issue, various sampling techniques have been employed to create synthetic samples to balance the training data. Even a single Machine Learning algorithm is capable of predicting churn but ensembles have gained popularity due to their robustness and better performance. Consequently, this research work has been experimented with various ensemble algorithms, which led us to the optimal model that combines the results from three ensembles i.e., Random Forests, Extremely Randomized Trees and Adaboost, to achieve better classification performance than any individual or ensemble algorithm. The results obtained by this model can be utilized by banks to make savvy business decisions and take strategic actions to prevent customer churn.