Comparasion of Support Vector Machine and Decision Tree Methods in Sentiment Analysis of Social Media X User Toward the Free Nutritious Meal Program

Figure Ilustration AI

FORMOSA NEWS - Medan - Support Vector Machine Algorithm Outperforms Decision Tree in Sentiment Analysis of Indonesia's Free Nutritious Meal Program. Indonesia's Free Nutritious Meal program, designed to curb nationwide stunting and improve student nutrition, has sparked extensive public conversation on social media. To objectively map public perception, computer science researchers Olinda Nathaniel Mendrofa, Deva Angriani, and Elvis Sastra Ompusunggu from Universitas Prima Indonesia conducted a comparative sentiment analysis on social media platform X. Published in August 2026 in the Formosa Journal of Computer and Information Science, the study evaluated two artificial intelligence algorithms: Support Vector Machine (SVM) and Decision Tree. The researchers discovered that the SVM algorithm achieved superior performance with an overall classification accuracy of 75.42 percent, outperforming Decision Tree's 72.08 percent. This research provides a data-driven framework enabling policymakers to monitor public feedback accurately and efficiently.

Stunting remains a critical public health challenge affecting long-term child development, cognitive readiness, and physical growth. According to World Health Organization estimates from 2024, approximately 150.2 million under-five children globally experience stunting. In Indonesia, the national stunting rate dropped to 19.8 percent according to the 2024 Indonesian Nutrition Status Survey. To accelerate stunting reduction under Presidential Regulation Number 72 of 2021, President Prabowo Subianto's administration introduced the Free Nutritious Meal program targeting 82.9 million beneficiaries, including toddlers, school students, and expectant mothers. Because social media platform X hosts over 24 million active users in Indonesia, it serves as a primary hub for real-time public commentary regarding this national policy.

To analyze public sentiment toward the meal initiative, the research team from Universitas Prima Indonesia implemented a machine learning workflow using Natural Language Processing techniques. Olinda Nathaniel Mendrofa and her co-authors gathered 1,200 public tweets between January 2025 and February 2026 using the Tweet Harvest scraping tool on Google Colab. The collected text data underwent extensive pre-processing, including text cleaning, case folding, tokenization, normalization, stopword removal, and word stemming.

Following data preparation, the authors applied Term Frequency-Inverse Document Frequency (TF-IDF) feature extraction to convert cleaned text into numerical matrix representations. This step allowed the classification models to weigh word significance computational across the dataset. The researchers divided the dataset using an 80:20 train-test ratio, allocating 960 sample tweets for model training and 240 tweets for independent testing. Model performance was rigorously evaluated via confusion matrix metrics measuring accuracy, precision, recall, and F1-Score.

The primary findings of the study reveal key insights regarding public sentiment and machine learning model effectiveness:

• Out of 1,200 labeled tweets, public feedback skewed predominantly negative with 526 negative entries (43.83 percent), followed by 477 positive entries (39.75 percent) and 197 neutral entries (16.42 percent).
  
• The Support Vector Machine algorithm outperformed Decision Tree across all metrics, recording 75.42 percent accuracy, 74.65 percent precision, 75.42 percent recall, and a 74.96 percent F1-Score.
  
• The Decision Tree model yielded lower performance metrics, achieving 72.08 percent accuracy, 73.33 percent precision, 72.08 percent recall, and a 72.59 percent F1-Score.
  
• SVM's optimal hyperplane boundary method proved more effective at separating high-dimensional, complex text data than Decision Tree's rule-based branching structure.
  
• Word cloud visualizations showed positive discussions focused on "children," "healthy," "smart," and "future," whereas negative comments centered on concerns regarding "budget," "food poisoning," and "program execution".
  

These findings offer practical benefits for public administration, educational policy, and computational data science. For government officials and the National Nutrition Agency, automated sentiment classification highlights operational concerns—such as food safety and budget allocation—that require immediate policy adjustment. For the technology sector, this research demonstrates that AI-driven text analytics can replace traditional, costly public opinion surveys with real-time digital monitoring. 

Author Profiles:
Olinda Nathaniel Mendrofa, B.Comp.Sc. is a computer science researcher at Universitas Prima Indonesia specializing in Data Science, Machine Learning, and Sentiment Analysis.
Deva Angriani, B.Comp.Sc. is a researcher at Universitas Prima Indonesia with expertise in Natural Language Processing, Text Mining, and Artificial Intelligence Applications.
Elvis Sastra Ompusunggu, M.Comp.Sc. is an academic and computer science researcher at Universitas Prima Indonesia specializing in Algorithm Analysis, Computational Modeling, and Information Systems.

Source
Mendrofa, O. N., Angriani, D., & Ompusunggu, E. S. (2026). Comparasion of Support Vector Machine and Decision Tree Methods in Sentiment Analysis of Social Media X User Toward the Free Nutritious Meal Program. Formosa Journal of Computer and Information Science (FJCIS), Vol. 5, No. 2, hlm. 229–242.
DOI: https://doi.org/10.55927/fjcis.v5i2.16668
URL: https://journal.formosapublisher.org/index.php/fjcis