<![CDATA[Rapid internet growth has accelerated e-commerce expansion and intensified competition among platforms. Because customers can move to competing platforms that better meet their needs, customer churn has become an important problem requiring classification based on historical customer behavior. This study aimed to evaluate the performance of the Random Forest Classifier combined with SMOTE and SMOTEENN resampling techniques for handling imbalanced e-commerce customer churn data. The research involved data cleaning, selection, transformation, and resampling, followed by Random Forest parameter tuning using GridSearchCV. The dataset was divided into 70% training data and 30% testing data, and performance was evaluated using accuracy, precision, recall, F1-score, confusion matrix, and AUC. Random Forest with SMOTE produced the best overall balance, achieving 96.3% accuracy, 87.8% precision, 87.1% recall, 87.4% F1-score, and 93% AUC. Random Forest with SMOTEENN achieved the highest recall of 91.5% and an AUC of 92%; however, its precision decreased to 66.2%, indicating more false-positive predictions. These findings show that SMOTE provides a more balanced classification performance for the evaluated dataset, whereas SMOTEENN prioritizes churn detection at the cost of precision. Practically, selecting a resampling strategy should reflect whether balanced performance or maximum churn detection is the primary operational priority in practical customer-retention decisions.]]>
Copyrights © 2023
