An Improved Random Forest Algorithm (ERFA) Utilizing an Unbalanced and Balanced Dataset to Predict Customer Churn in the Banking Sector

Customer churn, the phenomenon of customers discontinuing their relationship with a company, poses significant challenges for businesses in various industries, including the banking sector. In this paper, we address the problem of customer churn in the banking sector by selecting and applying four machine learning algorithms on the dataset of the U.S Bank in the case of Imbalance and Balance to predict the churn of customers from financial institutions such as banks and the reasons that drive them to do so. The algorithms were Logistics Regression, SVM, Neural network and Random Forest. The CRISP model was followed for the necessary processing and analysis, as it became clear after that the Random Forest algorithm provided better results in all cases. Consequently, this paper attempted to provide a better improvement to the results obtained from the default Random-Forest algorithm by understanding its working mechanism and adjusting its hyperparameters. Therefore, a mathematical model was proposed and built for the default Random Forest algorithm as a new version of it called ERFA. The final results of ERFA were positive through an increase in prediction accuracy from 85% to 86% in the case of the imbalanced dataset and from 89% to 91% in the case of balanced dataset. The outcomes of this study can help and motivate bank managers in general and in Yemen in particular to adopt this type of smart strategies to maintain their customers and provide them with a better experience. Furthermore, the mechanism established by this paper may help researchers who are interested in the field of developing smart business models.