Comparison of SVM and Random Forest for TikTok E10 Fuel Sentiment Analysis

One source of data that can be used to gauge public opinion on a specific public policy is social media. This study is to examine public opinion regarding the policy on the usage of a 10% bioethanol mix (E10) based on user comments on the TikTok platform. The sentiment analysis approach uses two classification algorithms Random Forest (RF) and Support Vector Machine (SVM). Pretreatment stages of data processing include tokenization, stemming, and lexicon-based techniques for identifying sentiment polarity. The Term Frequency–Inverse Document Frequency (TF-IDF) approach is used to extract features. The Synthetic Minority Over-sampling (SMOTE) technique was used to address class distribution imbalance in the data. Based on the test results, the accuracy achieved by the SVM and RF algorithms was 82.19% before applying SMOTE. Accuracy increased to 89.55% for SVM and 89.59% for RF after data balancing with SMOTE. Additionally, there was a more consistent improvement in precision, recall, and F1-score values. The findings of this study indicate that the use of SMOTE can improve the performance of classification models and reduce bias caused by class imbalance in sentiment analysis related to the E10 policy on social media.