Machine Learning Predictive Models Analysis on Telecommunications Service Churn Rate

Abstract

Customer churn frequently occurs in the telecommunications industry, which provides services and can be detrimental to companies. A predictive model can be useful in determining and analyzing the causes of churn actions taken by customers. This paper aims to analyze and implement machine learning models to predict churn actions using Kaggle data on customer churn. The models considered for this research include the XG Boost Classifier algorithm, Bernoulli Naïve Bayes, and Decision Tree algorithms. The research covers the steps of data preparation, cleaning, and transformation, exploratory data analysis (EDA), prediction model design, and analysis of accuracy, F1 Score, receiver operating characteristic (ROC) curve, and area under the ROC curve (AUC) score. The EDA results indicate that the contract type, length of tenure, monthly invoice, and total bill are the most influential features affecting churn actions. Among the models considered, the XG Boost Classifier algorithm achieved the highest accuracy and F1 score of 81.59% and 74.76%, respectively. However, in terms of efficiency, the Bernoulli Naïve Bayes and Decision Tree algorithms outperformed XG Boost, with AUC scores of 0.7469 and 0.7468, respectively.