Hybrid Machine Learning for Sentiment Analysis of Dana Application Reviews

This research evaluates user sentiment towards the Dana application on the Google Play Store, where in early 2024 1,000 reviews were collected. Of these reviews, 72% (720 reviews) were negative, while 28% (280 reviews) were positive. This situation arose because the Dana application was under maintenance during the dataset collection period. This research utilizes Support Vector Machine (SVM), Naive Bayes, and Hybrid methods for sentiment classification. The evaluation results show an accuracy of 92.62% for SVM, 88.62% for Naive Bayes, and 93.88% for Hybrid, where the Hybrid method shows the best performance in predicting user sentiment. This research makes an important contribution to the development of sentiment classification algorithms and provides insight for application developers to understand user perceptions during the repair period. It is hoped that the research results can help improve the quality of the Dana application and similar applications in the future. Highlights: Performance: The Hybrid Machine Learning method outperformed SVM and Naive Bayes, achieving the highest accuracy of 93.88%. User Feedback: Majority of reviews (72%) were negative due to application maintenance issues during the dataset collection. Future Development: Recommendations include using larger datasets and advanced NLP techniques like BERT for better sentiment analysis. Keywords: Dana App, Hybrid Machine Learning, Sentiment Analysis