Churn Prediction in a Pay-TV Company via Data Classification

Ilayda Ulku, Fadime Yuksektepe, Oznur Yilmaz, Merve Aktas, Nergiz Akbalik

OAI: oai:igi-global.com:266495 • DOI: 10.4018/IJAIML.2021010104

In data mining, if a data set is new to the literature, the study is comparing the existing algorithms and determining the most suitable algorithm. This study is an example of this by including many quantitative analysis. Real data was obtained from a Pay-TV Company in Turkey to predict the churn behavior of the customers. The attributes such as membership period, payment method, education status, and city information of customers were used in order to predict the customers' churn status. By applying attributes selection algorithms, the most important attributes are obtained. As a result, two datasets are proposed. While one of the datasets consists of all attributes, the other one just includes the selected attributes. Many different data classification algorithms were applied to these datasets by using WEKA software. The best method and the best dataset which has the best accuracy rate was proposed to the company. The company can predict the customers' churn status and contact the right group of people for a specific campaign with a proposed user-friendly prediction methodology.

Computer Science and Information Technology Artificial Intelligence Machine Learning