An interpretable machine learning Framework for customer churn Prediction: A case study in the telecommunications industry

Customer churn prediction has been gaining significant attention due to the increasing competition among mobile service providers. Machine learning algorithms are commonly used to predict churn; however, their performance can still be improved due to the complexity of customer data structure. Additionally, the lack of interpretability in their results leads to a lack of trust among managers. In this study, a step-by-step framework consisting of three layers is proposed to predict customer churn with high interpretability. The first layer utilizes data preprocessing techniques, the second layer proposes a novel classification model based on supervised and unsupervised algorithms, and the third layer uses evaluation criteria to improve interpretability. The proposed model outperforms existing models in both predictive and descriptive scores. The novelties of this paper lie in proposing a hybrid machine learning model for customer churn prediction and evaluating its interpretability using extracted indicators. Results demonstrate the superiority of clustered dataset versions of models over non-clustered versions, with KNN achieving a recall score of almost 99% for the first layer and the cluster decision tree achieving a 96% recall score for the second layer. Additionally, parameter sensitivity and stability are found to be effective interpretability evaluation metrics.