<![CDATA[hurn is the act by which a customer withdraws from service, including service provider-initiated churn and customer-initiated churn. Churn is a big challenge for companies, especially churn-prone enterprise sectors such as telecommunications. Churn can affect both revenue and reputation if occurs for negative reasons. This study aims to predict customer churn in a telecommunication company dataset, investigating the impact of various variables and classes on churn occurrences to inform strategic decision-making for businesses. The Support Vector Machine (SVM) model is employed, and dataset imbalance is addressed through oversampling techniques, specifically Synthetic Minority Over-sampling Technique (SMOTE) and random oversampling (ROS). Three SVM models are created with different training datasets (normal, SMOTE, ROS), yielding varying results. The normal dataset achieves the highest accuracy at 92%, outperforming SVM with ROS (89%) and SVM with SMOTE (87%). However, the normal dataset exhibits lower sensitivity compared to both oversampling techniques. The study identifies the cause of decreased accuracy in oversampling and low sensitivity in the normal dataset. The novelty of this research lies in testing the SVM model's ability to surpass the accuracy of previous models on the same dataset and in exploring the unique impact of oversampling in churn prediction.]]>
Copyrights © 2024
