MODEL TREE ENSEMBLE LEARNING FOR CHURN PREDICTION USING THEGRADIENT BOOST TREE AND EXTREME GRADIENT BOOST TREE METHODS

• Main Author: Ayu Lestari, Fitri
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/72960
• Institution: Institut Teknologi Bandung

Retaining customers to remain loyal to buy or use services is a challenging thing for companies. Customers who move from one company to another are called customer churn. Customer churn is a major concern for e-commerce companies because it has a negative impact on business growth. Customer churn can be predicted with customer data recorded during transactions. In this study, the authors will predict churn using the gradient boost (GBT) and extreme gradient boost (XGBT) methods as well as compare the two methods. Both of these methods were chosen because of their ability to handle complex classification problems and provide high prediction accuracy. In this research the author uses customer data of an electronic commerce that comes from open sources (kaggle). Both of these methods use an ensemble model, which combines trees to improve the accuracy and performance of predictions. To evaluate the performance of the model, the confusion matrix and k-fold cross validation were used. In addition, the ROC and AUC curves are also used to measure the model’s ability to predict customer churn. The results of this study show that the accuracy produced by the GBT method is 0.942 while the accuracy produced by the XGBT method is 0.964. The experimental results show that the two methods, GBT and XGBT, are able to provide good customer churn prediction results with a high score of accuracy. However, XGBT performs slightly better than GBT Boost in terms of churn prediction performance.