COMPARATIVE ANALYSIS OF BANK CUSTOMER CHURN PREDICTION USING LOGISTIC REGRESSION AND DECISION TREE

• Main Author: Gracia, Fiorenza
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/72949
• Institution: Institut Teknologi Bandung

Due to the rapid development of technology and innovation in the banking sector caused by the COVID-19 pandemic, the competition between banks is getting tougher. Besides gaining customers, maintaining a close relationship with existing customers is one of the company's obligations. Indeed, customer churn must be avoided. Therefore, analysis is needed to predict bank customer churn by considering several risk factors that become predictor variables, including credit score, gender, age, tenure, balance, number of products, credit card ownership, member activity rate, and estimated salary to the response variable which is whether the customer churn or not. This thesis compares Logistic Regression and Decision Tree models with the addition of the SMOTE-NC method to determine the best model for predicting bank customer churn. In this final project, the modeling used an open-source dataset from Kaggle. Based on the experiments, it suffices that the Decision Tree model with the application of the SMOTE-NC method with the ratio of the majority class 2:1 minority class was the best model for the data used compared to other models.