Sentiment Analysis on Vocational High School: A Machine Learning Approach

The increasing volume of user-generated content on digital platforms presents an opportunity to explore public perceptions of vocational education through sentiment analysis. In this research, sentiment classification is conducted on 976 public comments regarding SMKN 3 Bandung by implementing machine learning models including Support Vector Machine (SVM), Decision Tree, Random Forest, and XGBoost. The data were gathered from several digital platforms such as Instagram, TikTok, X (formerly Twitter), and Google Maps. The dataset is imbalanced, with fewer neutral and positive samples compared to negative ones, posing challenges for model training. To address this, both oversampling and careful preprocessing were applied, including case folding, tokenization, normalization, slang word replacement, stopword removal, and stemming. To measure the effectiveness of each model, evaluation was conducted using metrics such as precision, recall, accuracy, and F1 score. Results show that XGBoost achieved the highest performance on both automatically and manually labeled datasets, with the best result of 84% accuracy on manually labeled data. This highlights the strength of XGBoost and the value of high-quality manual annotation in sentiment classification for education-related content.