DESIGN OF CUSTOMER CHURN PREDICTION MODEL OF INDIHOME PT TELKOM INDONESIA USING DATA MINING

• Main Author: Damian, Hanzel
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70434
• Institution: Institut Teknologi Bandung

PT Telekomunikasi Indonesia (Telkom) is one of the biggest providers of information and communication technology services and telecommunication networks in Indonesia. PT Telkom provides services for business/corporate (B2B) and individual (B2C) based customers. One of the main B2C services provided by PT Telkom is IndiHome which provides fixed broadband service in the form of landline, internet, and interactive TV (cable TV). One of the main highlights of IndiHome's current problem is customer churn. Amidst the large number of competitors who provide fixed broadband services with various types of services and prices, IndiHome has a relatively higher churn rate of 2.1%. Moreover, there has been no effective handling yet by IndiHome to tackle this problem. To solve the problem, in this study was carried out the design of prediction model that aims to predict customer churn IndiHome customers so that IndiHome can minimize the churn rate it has. The prediction model is then packaged in the form of a simple application that can be used by IndiHome. The design of prediction model in this study refers to the methodology of CRISPDM (Cross Industry Standard Process for Data Mining). In this study, there are three alternative algorithms used to build prediction models, namely Artificial Neural Network, Random Forest, and Support Vector Machine. The three models generated from each algorithm will be evaluated based on several metrics such as accuracy, precision, recall, and F1 Score values and then will be selected one best model. The best model will be integrated with the application so that it can be used by IndiHome. The best models obtained in this study are Random Forest model with accuracy value of 86.75%, precision 93.08%, recall 79.39%, and F1 Score 85.69%. The model is integrated with an application in the form of GUI (Graphical User Interface) created using Python programming language.