An Explainable Random Forest Framework with Feature Engineering for Customer Churn in Telecommunications
Customer churn is a major problem for the telecommunications industry because it results in loss of revenue and increase in customer acquisition costs. Existing churn prediction methods mainly focus on the predictive accuracy and lack of explanations for the predictions. The study proposed an intelligent customer churn prediction framework combining predictive modeling and explainable artificial intelligence techniques to facilitate customer retention decision making. The objectives were to preprocess and prepare the customer data, engineer informative features, tackle class imbalance, develop and simulate a predictive model and provide explainable churn predictions. The dataset used in the study was Telco Customer Churn Dataset from Kaggle. Data preprocessing, feature engineering, feature scaling and class balancing using Synthetic Minority Oversampling Technique (SMOTE) were applied. A Random Forest classifier was trained as predictive engine and used feature importance analysis and SHapley Additive exPlanations (SHAP) to provide global and local explanations of model predictions. The developed model attained the accuracy of 79% with the precision of 0.6148, recall of 0.6192, F1-score of 0.6170, and receiver operating characteristic area under the curve (ROC-AUC) score of 0.8473. The results of the simulation showed practical value of the framework in finding customers that are likely to churn. The study concluded that the proposed framework is effective in predicting customer churn, while at the same time providing interpretable predictions. Telecommunication companies should embrace explainable predictive systems for proactive customer retention strategies.
Download Paper