<![CDATA[The ability to anticipate unsubscribed customers is a challenge in the competitive banking industry, where it is more efficient to retain customers than to attract new ones. The purpose of this study is to improve the effectiveness of churn prediction by overcoming data imbalances using SMOTE (Synthetic Minority Oversampling Technique) and Random Over-sampling. The data set used consists of 10. 000 bank customer data, with 12 important attributes, including churn indicators as targets. The machine learning algorithms used are Random Forest and Neive Bayes, evaluated based on accuracy, precision, recall, and F1 scores. The results of the experiment showed that the highest accuracy of 87.13% could be achieved with the Random Forest algorithm without using the oversampling method, but its effectiveness in detecting churn customers was slightly limited. The use of SMOTE and Random Over-sampling methods has improved the model's performance in identifying churn patterns, although it has led to a decrease in accuracy to 86.20% for Random Over-sampling and 81.47% for SMOTE. Nevertheless, the Neive Bayes algorithm showed the best accuracy rate of 79.20% without oversampling, although it was still slightly lacking in optimal churn handling. The study underscores the importance of using oversampling methods to improve prediction balance in minority classes, which is often overlooked in conventional models. It is hoped that the results of this research can be used as a guide in improving strategies to maintain customer trust that are more up-to-date and efficient.]]>
