Abstract
Penggunaan dompet digital yang terus meningkat menghasilkan banyak ulasan pengguna yang dapat dimanfaatkan untuk mengevaluasi kualitas layanan. Penelitian ini bertujuan meningkatkan akurasi klasifikasi sentimen pengguna dompet digital menggunakan metode Stacking Ensemble Machine Learning yang mengombinasikan Support Vector Machine (SVM), Multinomial Naïve Bayes (MNB), dan AdaBoost dengan Logistic Regression sebagai meta-learner. Data ulasan diproses melalui tahapan text preprocessing meliputi case folding, cleaning, tokenizing, stopword removal, stemming, dan pembobotan fitur menggunakan TF-IDF. Penyeimbangan data dilakukan dengan SMOTE, sedangkan evaluasi model menggunakan 5-Fold Cross-Validation. Hasil penelitian menunjukkan bahwa model Stacking Ensemble memperoleh akurasi rata-rata 80,55%, lebih tinggi dibandingkan algoritma dasar. Evaluasi menggunakan Confusion Matrix, Classification Report, dan ROC Curve juga menunjukkan peningkatan nilai precision, recall, F1-score, dan kemampuan diskriminasi model. Hasil ini menunjukkan bahwa pendekatan Stacking Ensemble Machine Learning efektif untuk meningkatkan akurasi klasifikasi sentimen pengguna dompet digital serta mendukung evaluasi kualitas layanan berbasis opini pengguna. The rapid growth of digital wallet usage has generated a large volume of user reviews that can be utilized to evaluate service quality. This study aims to improve the accuracy of digital wallet user sentiment classification using a Stacking Ensemble Machine Learning approach that combines Support Vector Machine (SVM), Multinomial Naïve Bayes (MNB), and AdaBoost with Logistic Regression as the meta-learner. User reviews were processed through text preprocessing stages, including case folding, text cleaning, tokenization, stopword removal, stemming, and TF-IDF feature weighting. Synthetic Minority Over-sampling Technique (SMOTE) was employed to address class imbalance, while model performance was evaluated using 5-Fold Cross-Validation. The experimental results show that the proposed Stacking Ensemble model achieved an average accuracy of 80.55%, outperforming the individual base learners. Furthermore, evaluations based on the Confusion Matrix, Classification Report, and Receiver Operating Characteristic (ROC) Curve demonstrated improvements in precision, recall, F1-score, and the model's discriminative capability. These findings indicate that the proposed Stacking Ensemble Machine Learning approach is effective in improving the accuracy of digital wallet user sentiment classification and can serve as a reliable tool for supporting service quality evaluation based on user opinions.
