A novel systematic forest classifier for detecting client churn in the telecommunications industry

Predicting client churn in the telecommunications industry is a significant challenge due to the high dimensionality and imbalance of its datasets. Conventional models like traditional decision trees (DTs) are not often able to deal with these intricacies, resulting in suboptimal performance. Furthermore, DTs are prone to overfitting and may require more effective capture of complex feature interactions, despite their popularity due to their simplicity and interpretability. Therefore, this research addresses these challenges by introducing a novel systematic forest (SF) classifier that systematically evaluates and selects attributes, thereby balancing the complexity and interpretability of the model. SF design enhances the understanding of feature interactions and provides a more nuanced approach to discovering client attrition. We evaluated its performance in predicting client churn in the telecommunications industry and compared it with a traditional decision tree model. This is utilizing the preprocessed benchmark Cell2Cell dataset, employing a combination of KNN imputation, MMADN with min-max normalization, and SMOTE Tomek resampling. Performance evaluation metrics include recall, accuracy, precision, F1 score, and ROC-AUC. With accuracy of 80%, precision and recall of 0.80, and an F1 score of 0.79, the SF performed better than the DT. The results propound that the SF model’s systematic approach advances attribute interaction analysis and offers a more robust prediction capability compared to the DT model.