ANALYSIS OF CUSTOMER CHURN PREDICTION USING LOGISTIC REGRESSION, -NEAREST NEIGHBORS, DECISION TREE AND RANDOM FOREST ALGORITHMS

Abstract

Customer churn predictions (CCPs) and their comprehensive analysis have become prevalent in the global telecom industry over the last five years, driven by advancements in machine learning (ML) technologies. In addition, AI (artificial intelligence) and ML-based predictive methods are currently employed for CCP applications to enhance customer retention. This predictive CCP methodology streamlines customer management processes and ensures sustainable profit growth. The machine learning models focus on identifying features derived from data that is rich in various types of information. This study analyzes CCP for a specific telecom company’s customer dataset using ML methods such as logistic regression (L.R.), -nearest neighbor (-NN), decision tree (D.T.), and random forest (R.F.). The UCI Iranian telecom churn dataset was utilized, and the influence of potential factors leading to customer churn was also considered. Results show that the tuned RF method yielded the best outcomes, with churn tendency analysis achieving a higher AUC score at 0.9042 with the accuracy of 0.9562. The most important feature of the dataset affecting the customer churn was identified as complains whereas the least important feature happened to be tariff plan.