A Comparative Study : Predicting Customer Churn in Banking Using Logistic Regression & Random Forest

Abstract

This research explores the prediction of bank customer churn using machine learning techniques. The dataset used includes various customer features such as demographics, transaction history, and interactions with the bank. After performing exploratory data analysis (EDA) and pre-processing, two machine learning models were applied: Logistic Regression and Random Forest. The EDA results showed that factors such as number of transactions, total transaction value, and credit utilization rate were correlated with the likelihood of churn. Pre-processing included handling categorical data, removing irrelevant features, and dividing the data into training and testing sets. The Logistic Regression model achieved 84% accuracy on training data and 83.9% on testing data, but showed poor performance in terms of recall and F1-score for the "Attracted Customer” class. In contrast, the Random Forest model showed excellent performance with 100% accuracy on both datasets, as well as perfect precision, recall, and F1-score values for both classes. In conclusion, the Random Forest model was selected as the best model to predict bank customer churn. These findings can help banks identify customers at risk of churn and develop effective retention strategies.