Optimasi Support Vector Machine Menggunakan Feature Selection Chi-Square untuk Klasifikasi Sentimen Program Makan Siang Gratis

The Free Nutritious Lunch program policy has triggered massive public discourse on social media X, reflecting diverse public perceptions toward government policy effectiveness. This study aims to optimize the performance of the Support Vector Machine (SVM) algorithm by implementing Chi-Square Feature Selection to address high data dimensionality and noise challenges in social media text. A dataset of 10,524 tweets was acquired and processed through preprocessing, TF-IDF weighting, and lexicon-based automatic labeling. The results show that Chi-Square feature selection integration successfully reduced dimensions from 16,394 to the 1,000 best features without degrading accuracy. The linear kernel SVM model achieved an optimal accuracy rate of 91.12%. However, this study identifies that this high accuracy is heavily influenced by the dominance of the positive class, whereas performance on the negative and neutral classes remains limited due to data imbalance. Overall, feature optimization proved to increase computational efficiency while maintaining accuracy stability in mapping public responses to strategic national policies.