Klasifikasi Sentimen Ulasan E-Wallet menggunakan TF-IDF dan Random Forest dengan Penyeimbangan Data SMOTE

Abstract

The increasing use of e-wallets in Indonesia highlights the need to understand user perceptions automatically and efficiently. One valuable data source is user reviews from the Google Play Store. This study aims to classify sentiment toward three major e-wallets, such as GoPay, OVO, and DANA to support service improvement. A quantitative approach is used with a machine learning-based classification method. A total of 30,000 reviews (10,000 per application) were collected using the google-play-scraper library. The data were processed through several stages: preprocessing (labeling, stopword removal, tokenization, and stemming), feature extraction using TF-IDF, data balancing with SMOTE, and classification with the Random Forest algorithm. Our findings show that the combination of Random Forest and SMOTE significantly improves model performance. Accuracy reached 90% (GoPay), 90% (OVO), and 87% (DANA). Precision, recall, and weighted f1-score were 90%, 89%, and 89% for GoPay; 90%, 90%, and 90% for OVO; and 88%, 88%, and 88% for DANA. WordCloud visualizations further support the findings by highlighting dominant words in each sentiment, such as “good,” “help,” and “lost.” Overall, the integration of TF-IDF, SMOTE, and Random Forest is proven effective and reliable for sentiment classification across the three e-wallet platforms.