THE EFFECT OF IMPERFECTIVE DATA SAMPLING METHOD ON SUPPORT VECTOR MACHINE ACCURACY

Abstract

Sentiment analysis is used to understand the direction of public opinion, but problems arise due to the unbalanced distribution of sentiment data, where one class dominates. This imbalance causes classification models such as Support Vector Machine (SVM) to be biased towards the majority class, which results in decreased accuracy and generalizability of the model. This study aims to assess the effectiveness of two data balancing techniques, namely, SVM-SMOTE, and ADASYN, in improving SVM performance. The research data was taken from social media platform X (Twitter), and testing was conducted using the K-Fold Cross Validation method (K=2, 5, and 10) using evaluation metrics such as accuracy, precision, recall, and F1-score. The results show that without data balancing, the SVM model can only achieve an average accuracy of 76.34% and F1-score of 62.38%, which reflects the weakness in recognizing minority classes. The application of the two balancing methods successfully improved the model performance. ADASYN increased the F1-score to 67.94%, while SVM-SMOTE showed the most optimal results with 82.4% accuracy and 74.02% F1-score. These findings indicate that SVM-SMOTE is the most effective technique in handling data imbalance and improving sentiment classification accuracy equally.