Churn Prediction in Telecom Industry Using Machine Learning Algorithms with K-Best and Principal Component Analysis

Abstract

The main feature in customer relationship management systems is customer churn prediction. Since customer churn directly impact the revenue of companies, finding factors and taking necessary actions are important. Performance of the model is measured using the Area Under Curve (AUC). The model experimented four machine learning algorithms: Extreme Gradient Boosting (XGBoost), Random Forest Classifier (RFC), Logistic Regression, and Support Vector Classifier (SVC). All the four models are applied on standard scalar normalized data as well on outputs of feature extraction model viz. Principal Component Analysis (PCA) and feature selection model viz. Select K-Best method. Reduction of fourteen to four features were achieved using Select K-Best and PCA. XGBoost algorithm gave better results than the other algorithms on Select K-Best outputs than on PCA outputs. Some of the advantages of XGBoost algorithm is its scalability, ability to handle missing values, and avoid overfitting.