Customer Churn Prediction Based on the Decision Tree and Random Forest Model

Abstract

The rate at which customers discontinue utilizing a company's services during a predetermined time period is known as the customer churn rate, also known as the attrition rate. Hence, developing a prediction model to predict the potential churn customers will generate an early alert for the company to provide them with better service. This study is divided into two main parts: dealing with a dataset about customer behaviors in a bank and building churn prediction models using machine learning algorithms. The data preprocessing part includes dataset description and some adjustments on original dataset to make it accessible for analysis, including deleting unimportant feature and adjusting feature names. Then the study apportions the modified dataset into train set and test set with an 80-20 split. Next, the study imports two kinds of machine learning algorithms, random forest classifier and decision tree classifier, to build churn prediction models. In each model, the study first performs feature selections and visualizes feature importance in bar graphs. Then the study tests each model on testing set and visualizes model performances using confusion matrices and accuracy scores. The results show that both models get most predictions correct while random forest model has a better performance due to its higher accuracy score of 91%.