Forecasting Customer Churn in the Telecommunications Industry

Abstract

Data mining is a broad field that helps the company to combine statistics, databases, machine learning, and artificial intelligence. As the size of the company grows, so do such situations, making it impossible for a normal information system to manage such perilous scenarios. Due to this companies face significant income loss since customers are leaving the firm for unexplained reasons. It is well acknowledged that acquiring new clients is more cash intensive than maintaining existing ones and hence customer management is critically impacted by customer churn, which happens when a customer decides he no longer wants to keep in touch with the company. Traditional market research methodologies are challenging to support the churn problem. There is still much potential for improvement in churn forecast accuracy despite the development of several churn prediction tools that look at hundreds of parameters. Ultimately, this research will aid in the analysis of consumer behavior and the categorization of whether or not a client is churning through the use of a variety of data mining approaches to predict customer churn. Using a data set available on Kaggle's website, this study tested multiple classifiers on the problem of predicting customers' propensity to leave a company. In this study, we utilized Kaggle's online data set to predict customer churn behavior using several classifiers, including Random Forest, Logistic, J48, Stacking, ADA Boost, Decision Table, and Logit Boost, and observed that our model achieved 93.55 percent accuracy.