Authors - Josephine Florencia Chan, Anderes Gui, Riki, Huynh Trong Thua, Nguyen Minh Tuan, Chau Van Van Abstract - Losing customer in the telecommunication may lead to significant financial losses. Machine learning approaches have shown promising potential for predicting churn, but many studies still focus primarily on Accuracy, which can be misleading when using an imbalanced dataset. This study compares three ma-chine learning algorithms: Logistic Regression, Linear Support Vector Machine (SVM), and Decision Tree. The goal is to determine which algorithms prioritizes Recall. The Iranian Churn dataset was used for the experiment; this dataset con-sists of 3151 customer records with 14 behavioral and demographic attributes. This study used an 80:20 train-test split with standardized features, and model performance was evaluated based on Recall, F1-score, Precision, Specificity, and Accuracy. The Decision Tree model achieved the highest Recall, while Logistic Regression and Linear SVM showed slightly lower Recall but similar Accuracy. These results suggest that for small and structured customer datasets, simpler or appropriately constrained models may perform effectively while prioritizing the identification of churners. Model selection should consider dataset characteris-tics. Prioritizing Recall over Accuracy can also help guide effective customer retention strategies.
Open Zoom