Machine Learning-based Analysis and Prediction of Telecoms Customer Churn

Today, in the crowded and competitive telecom market, countless telecom companies suffer from customer churn. To help telecom companies predict the potential churn rate of their customers, this paper proposes to apply machine learning algorithms to construct evaluation models. Specifically, the experiment consists of five stages: data preprocessing, feature engineering, data segmentation, model building and evaluation. Missing values are dealt with in the data preprocessing and feature engineering phases. Transformation of variables is also performed. A 70% training set and a 30% testing set are then created. The unbalance in the training set is subsequently addressed using the Synthetic Minority Oversampling Technique (SMOTE). Support vector machine (SVM) and random forest (RF) models are then constructed to predict customer churn. Finally, the prediction results of the testing set are evaluated by comparing the confusion matrix and Receiver Operating Characteristic (ROC) curves. Eventually, with accuracy, F1 score, accuracy, and Area Under Curve (AUC) values of 0.80, 0.76, 0.87, and 0.93, individually, the RF method outperforms the SVM. The experimental findings demonstrate that the RF method is capable of efficiently analyzing customer turnover data and making accurate predictions.