A customer churn prediction in telecom service provider using machine learning techniques

Abstract

With the rapid growth of digital and technology, a big amount of data is generated on a daily basis from customers, and customer churn is causing a great concern in this highly competitive service sector, especially the telecom service providers. Investment in acquiring new customers is more expensive than retaining existing ones; therefore, business analysts and customer relationship management systems need to assess customer churn based on the available data. This research proposed a churn prediction model that used classification and dataset extracted from the data warehouse of the telecom service provider in Thailand. RapidMiner software was used to implement the following techniques: the decision tree algorithm, the random forest algorithm, the gradient boosted algorithm, and the neural networks algorithm. Confusion metrics in terms of accuracy, precision, recall, and receiving operating characteristics (ROC) area were used to evaluate the suggested churn prediction model. The experiment found that neural networks had the highest prediction for new data input of AUC=0.866 and that the decision tree performed the second best with 87.58% accuracy and had the fastest speed. Implication of Thematic Paper: The results can be used to create a decision support system regarding classification approaches for turbulent customers. Further studies may include techniques such as pre-preparation with more data elements, number of algorithms, and optimization of data analysis. Configuring each algorithm to identify the model that will be optimized for more results. This can increase the efficiency of choosing the most suitable method of churn predicting and working with the company's customer retention campaign.