Bank Churn Prediction Using Random Forest and Logistic Regression

In the banking industry, customer churn is a growing problem. Solving this problem effectively and choosing the appropriate forecasting model is important. To avoid such problems and select an appropriate model, in this paper, random forest and logistic regression models are used to predict bank churn based on a specific data set. Different situations are set to evaluate the models. During prediction, the parameters of the model and variables are changed slightly for comparison. The accuracy, recall, precision, and stability of models are compared. The accuracy of random forest is about 86%, nearly 3 points higher than logistic regression. After removing the least correlative factor, the accuracy of the random forest remained almost unchanged, while logistic regression had a 4-point decline. Fluctuation brought by removing the least correlative variable is smaller in the random forest which means better stability. Though this study has shown a random forest’s better performance, removing the least correlative factor leads to a decline in both models. This is contradicted by the author’s hypothesis. Hence, further study with enough features will be a good way to compare more about these two models in bank churn prediction.