COMPARATIVE MACHINE LEARNING ALGORITHMS FOR YOUTUBE SENTIMENT ANALYSIS ON DPR DEMONSTRATION 2025 USING LEXICON

Abstract

The high volume of public comments on YouTube regarding the DPR Demonstrasion August 2025, which reached 43,910 raw data, presents a significant challenge in conducting efficient sentiment analysis. Time and cost limitations in manual labeling for large-scale datasets are a major obstacle in the development of predictive models. This study aims to address this problem by proposing a hybrid approach that integrates Lexicon-Based auto-labeling with a comparative evaluation of five Machine Learning algorithms. The research methodology included a text preprocessing stage that generated 40,097 unique comments, feature extraction using TF-IDF, and data sharing with an 80:20 ratio. The performance of the Support Vector Machine algorithm was comprehensively compared to Random Forest, Decision Tree, K-Nearest Neighbors, and Naive Bayes. The results of the experiment showed that the SVM model recorded the most superior performance with an accuracy of 96.5% and a weighted F1-Score of 0.966. This score significantly outperformed other benchmarking algorithms, where Random Forest came in second place with 89.2% accuracy, followed by Decision Tree at 85.6%, KNN at 84.6%, and Naive Bayes at the lowest with 84.0%. These findings validate that the integration of Lexicon-Based labeling with SVM classification is a highly accurate, robust, and efficient solution for handling sentiment analysis on large-scale social media data in Indonesia.