Comparative Analysis of Naive Bayes and SVM for Improved Emotion Classification on Social Media

identifying emotions such as happy, angry, sad, and fear. However, Indonesian text processing faces challenges due to language complexity and slang. This research aims to compare Naive Bayes and SVM models, focusing on evaluating the impact of preprocessing, feature extraction, and parameter optimization to improve emotion classification. The dataset was collected from API X using crawling techniques and manually annotated by six annotators. The training process used full and half preprocessing datasets with TF-IDF, BoW, and Word2Vec feature extraction. Naive Bayes and SVM models were evaluated using accuracy, precision, recall, and F1 score. Our results show that full preprocessing improves accuracy, with TF-IDF + BoW achieving 78.01% with SVM and outperforming Naïve Bayes at 75.53%. The results classify emotions into four classes: happy, sad, angry, and fear. This study demonstrates the value of preprocessing and feature selection to deal with slang and complexity in Indonesian texts. These results provide insights for developing optimal emotion classification models and offer applications in sentiment analysis, social media monitoring, and mental health detection.