Analisis Performa Metode Machine Learning dalam Mengidentifikasi Penyebab Ulasan Rating Satu Aplikasi MyBluebird

Abstract

This study addresses the increasing prevalence of negative user reviews for the MyBluebird ride-hailing application, focusing on the identification and classification of the main causes of one-star ratings. The research aims to compare the effectiveness of Support Vector Machine, Random Forest, and Naïve Bayes algorithms in classifying user complaints. Employing a quantitative experimental approach, the study utilizes a dataset of 1,399 one-star reviews collected purposively from Google Play Store. Data preprocessing includes cleaning, tokenization, and feature extraction using TF-IDF. The classification models are evaluated using accuracy, precision, recall, and F1-score metrics. Results indicate that Random Forest achieves the highest accuracy (90%), outperforming the other algorithms, with bugs/errors as the most frequent complaint, followed by driver performance, other issues, and price. The study concludes that machine learning-based classification can effectively map user dissatisfaction, though data imbalance remains a limitation. Future research should apply data balancing techniques and expand the dataset for broader generalization. Practical implications suggest that developers can utilize automated classification to improve service quality and address user needs more efficient.