<![CDATA[The rapid development of digital technology has encouraged the adoption of mobile banking applications in Indonesia, but it has also led to an increase in user complaints and reviews regarding performance and ease of use. This study aims to conduct a comparative analysis of the performance of Machine Learning, Deep Learning, and Transformer (IndoBERT) models in classifying the sentiment of user reviews of Indonesian-language mobile banking applications. Data was collected through web scraping from the Google Play Store on ten leading banking applications in Indonesia with a total of 200,000 reviews. After going through the preprocessing stages of cleaning, normalisation, tokenisation, and stemming, automatic labelling was carried out based on ratings into three sentiment classes: positive, neutral, and negative. Machine learning models (Naïve Bayes, Logistic Regression, Random Forest, and SVM) were built using TF-IDF feature representation, while deep learning models (LSTM, Bi-LSTM, GRU, and CNN) utilised 128-dimensional word embeddings. The Transformer-based IndoBERT model was fine-tuned with a sequence classification configuration. The evaluation used accuracy, precision, recall, and weighted F1-score metrics, accompanied by an analysis of training and testing time efficiency. The results show that the Bi-LSTM model performs best with an accuracy of 83.47% and an F1-score of 80.78%, followed by CNN (83.11%) and SVM (82.85%), while IndoBERT records an accuracy of 81.73% with a precision of 76.96%. In terms of efficiency, Logistic Regression showed an optimal balance between accuracy and training time (27.7 seconds), while deep learning and transformer models required higher computational resources. This study emphasises the importance of model selection based on requirements, between maximum accuracy and computational efficiency, and enriches the literature on Indonesian sentiment analysis in the domain of digital financial services.]]>
Copyrights © 2026
