Sentiment Classification of Indonesia’s Free Nutritious Meal Policy on X Using Machine Learning Algorithms

The Free Nutritious Meal (Makan Bergizi Gratis, MBG) program has generated extensive discussion on X, particularly regarding children’s nutrition, budget allocation, and program implementation. This study aims to classify public sentiment toward the program and compare the performance of machine learning algorithms for MBG-related sentiment classification. The study applied the Knowledge Discovery in Databases framework, covering data collection, preprocessing, transformation, labeling, classification, and evaluation. After data cleaning, 10,001 tweets collected from X using the keyword “Makan Bergizi Gratis” were retained for analysis. Initial sentiment labels were generated using a lexicon-based approach and manually validated on 2,500 tweets. The validated subset was evaluated using 10-fold cross-validation. Naïve Bayes, Decision Tree, and Support Vector Machine were compared using accuracy, precision, recall, F1-score, and confusion matrices. Under the implemented default settings, Naïve Bayes achieved an accuracy of 73.64% ± 2.20% and a macro F1-score of 72.82%. The selected model was then applied to the remaining 7,501 tweets for model-based sentiment prediction. The results show that machine learning-based sentiment analysis can support the monitoring of public responses to policy, although model-based predictions should not be interpreted as definitive ground-truth public sentiment.