<![CDATA[Sentiment analysis of user reviews on the Mobile JKN application plays an important role in identifying user satisfaction levels and obstacles encountered when accessing BPJS Kesehatan’s digital services. This study aims to evaluate the performance of three classification algorithms Naïve Bayes, KNN, and SVM optimized using the SMOTE approach to address data imbalance. A total of 19,186 user reviews were used, which underwent preprocessing, TF-IDF weighting, and an 80:20 data split, resulting in 15,348 training data and 3,838 testing data. Each model was evaluated before and after SMOTE application to assess its impact on model performance. The results show that SMOTE improved model sensitivity toward minority classes without significantly reducing accuracy. The Naïve Bayes model achieved 93.56 % accuracy before SMOTE and 92.05 % after, KNN improved from 71.20 % to 87.96 %, while SVM maintained the highest performance with 93 % accuracy and an F1-score of 0.91–0.95. The highest AUC value was obtained by SVM (0.98), followed by Naïve Bayes (0.97) and KNN (0.93). Sentiment classification results revealed that most user reviews were positive, reflecting high satisfaction with the Mobile JKN service, while negative reviews were primarily related to technical issues. Overall, the SMOTE-based SVM model proved to be the most effective algorithm for sentiment analysis and can serve as a foundation for evaluating and improving the quality of BPJS Kesehatan’s digital services.]]>
Copyrights © 2026
