ROS-Enhanced Sentiment Analysis of Social Media Discourse on Indonesia's 2025 Budget Efficiency Policy Using Support Vector Machine

The 2025 State Budget (APBN) is a key instrument for the Indonesian government to sustain economic stability and support national development. Persistent challenges such as deficits, low absorption, and inefficiency prompted the issuance of Presidential Instruction No. 1 of 2025 on budget efficiency. While the policy seeks to reduce non‑productive expenditures, it has also raised concerns about declining public service quality, sparking public debate. This study examines public sentiment toward the policy by analyzing 546 comments from the social media platform X, categorized into positive and negative classes. The methodological framework includes text preprocessing, TF‑IDF weighting, and sentiment classification using Support Vector Machine (SVM). To address class imbalance, Random Oversampling (ROS) was applied to the training dataset under an 80:20 train-test split. Results show that SVM achieved strong performance on the original dataset (F1 Score 83.12%, Accuracy 87.04%), and its metrics remained relatively stable after ROS (F1 Score 81.64%, Accuracy 85.19%), indicating robustness to imbalance. These findings confirm that SVM provides consistent performance for sentiment analysis in policy-related discourse, while highlighting that oversampling yields minimal benefit. The study offers empirical insights into public responses to budget efficiency measures and methodological approaches for handling imbalanced data.