<![CDATA[Social media has become a strategic platform in conveying public opinion, especially at the moment of the Regional Head Election (Pilkada). The large amount of opinion data produced opens up opportunities for the application of sentiment analysis to map public perception. One of the main challenges in the classification of sentiment is the imbalance of distribution between classes, which can degrade the accuracy of the model, especially in recognizing minority classes. This study aims to analyze the impact of the application of data balancing techniques on the performance of the 2024 East Java Regional Election sentiment classification model using the Random Forest algorithm. The series of processes in the study include data preprocessing, manual sentiment labeling, text preprocessing, word weighting with TF-IDF, and model training on three data ratios, namely 90:10, 80:20, and 70:30. Each ratio was tested in three scenarios, namely no balancing (baseline), undersampling using the Tomek Links method, and oversampling using Borderline-SMOTE. Of all scenarios, Borderline-SMOTE gave the highest accuracy of 82.40% at an 80:20 ratio, an increase of 2.19% compared to the unbalanced condition at the same ratio. These results show that data balancing is able to improve the performance of the model in classifying sentiment more proportionally.]]>
Copyrights © 2026
