<![CDATA[Abstract User reviews of the Akulaku application on Google Play Store contain important information regarding user satisfaction and complaints. This study compares the performance of Naïve Bayes and Random Forest algorithms in classifying sentiment into three classes: positive, negative, and neutral. A total of 2,000 Indonesian-language reviews were collected using web scraping techniques. Data were processed through case folding, cleaning, normalization, stopword removal, tokenizing, and stemming. Labeling was based on user ratings. After preprocessing, 1,953 data points remained with an imbalanced distribution; SMOTE was applied to balance each class to 1,114 samples. TF-IDF was used for feature weighting with an 80:20 train-test split. Results showed Random Forest achieved 83% accuracy, while Naïve Bayes reached 80%. However, Naïve Bayes outperformed in precision, recall, and F1-score with macro averages of 0.59, 0.70, and 0.60, compared to Random Forest at 0.52, 0.55, and 0.54. Based on these results, the choice of the best algorithm depends on the specific needs. If the priority is overall accuracy, Random Forest is more recommended however, if the priority is balanced performance across sentiment classes, Naïve Bayes performs better.]]>
Copyrights © 2026
