CUSTOMER CHURN PREDICTION MODEL: A CASE OF THE TELECOMMUNICATION MARKET

Abstract

The telecommunications market is well developed but is characterized by
oversaturation and high levels of competition. Based on this, the urgent problem
is to retain customers and predict the outflow of customer base by switching
subscribers to the services of competitors. Data Science technologies and
data mining methodology create significant opportunities for companies that
implement data analysis and modeling for development of customer churn
prediction models. The research goals are to compare different approaches and
methods for customer churn prediction and construct different Data Science
models to classify customers according to the probability of their churn from
the company's client base and predict potential customers who could stop to
use the company's services. On the example of one of the leading Ukrainian
telecommunication companies, the article presents the results of different
classification models, such as C5.0, KNN, Neural Net, Ensemble, Random Tree,
Neural Net Ensemble, etc. All models are prepared in IBM SPSS Modeler and
have a high level of quality (the overall accuracy and AUC ROC are more than
90%). So, the research proves the possibility and feasibility of using models in the
further classification of customers to predict customer loyalty to the company
and minimize consumer’s churn. The key factors influencing the customer
churn are identified and form a basis for future prediction of customer outflow
and optimization of company’s services. Implementation of customer churn
prediction models will help to maintain customer loyalty, reduce customer
outflow and increase business results.