Evaluating Machine Learning Models for Sentiment Analysis on Indonesian Social Media

This study evaluates the performance of machine learning algorithms for sentiment analysis on Indonesian social media, driven by the increasing role of digital platforms in public discourse during significant corporate and political events. The research focuses on analyzing conversations from Indonesian users on X (formerly Twitter) concerning the Pertamina corruption case to uncover dominant public perspectives and compare algorithmic efficacy. Employing a robust methodology, the study applied and compared three machine learning algorithms Naive Bayes, Support Vector Machine (SVM), and Random Forest with extensive preprocessing techniques including TF-IDF for feature extraction and SMOTE for addressing class imbalance. Performance was rigorously assessed using metrics such as accuracy, precision, recall, and F1-Score. Findings reveal that the Random Forest model achieved the highest accuracy (88.9%), closely followed by SVM (88.6%), significantly outperforming Naive Bayes (83.3%). This research not only identifies a dominant negative public sentiment in the analyzed discourse but also validates an effective computational model for sentiment analysis tasks in similar socio-political contexts, advancing the application of machine learning in opinion mining.