Implementasi Logistic Regression dan SMOTE dalam Analisis Sentimen Ulasan Wondr by BNI

Abstract

Rapid innovation in Indonesia’s digital banking, evidences by digital transactions reaching Rp7,492.93 trillion by September 2024 and the launch of Wondr by BNI by PT. Bank Negara Indonesia, which has 5.3 million active users and 397 thousand reviews on Google Play Store by June 2025, presents a challenge for manual sentiment analysis of reviews due to its inefficiency. This study addresses this issue by employing a machine learning approach, utilizing the Logistic Regression algorithm for sentiment analysis. A total of 8000 review data from Kaggle were used, with sentiment labeled based on rating scores (1-3 negative, 4-5 positive). The methodology included data preprocessing, feature weighting using Term Frequency-Inverse Document Frequency (TF-IDF), and balancing training data with Synthetic Minority Oversampling (SMOTE). The Logistic Regression model was trained after parameter optimization via grid search, yielding the optimal combination of C=1, penalty=’l2’, and solver=’newton-cg’. Evaluation using a confusion matrix revealed an overall accuracy of 93.54%. For negative sentiment, the model achieved 71.89% precision, 91.5% recall, and 80.52% F1-score, while for positive sentiment, it reached 98.48% precision, 93.89% recall, and 96.13% F1-score.