Multiclass Sentiment Analysis of Electric Vehicle Incentive Policies Using IndoBERT and DeBERTa Algorithms

The electric vehicle (EV) incentive policy in Indonesia has generated various public reactions, particularly on social media platforms. This study aims to classify public sentiment using the IndoBERT and DeBERTa transformer models. A total of 6,758 comments were collected from YouTube, filtered, preprocessed, and labeled into three sentiment categories: positive, negative, and neutral. From this, 1,711 clean data points were used and analyzed in two phases: before and after applying the Random Oversampling technique to address class imbalance. Model performance was evaluated using accuracy, precision, recall, F1-score, and training time. In the initial phase, IndoBERT achieved 96% accuracy with 603.71 seconds of training time, while DeBERTa reached 93% in 439.19 seconds. After balancing and applying 5-Fold Cross Validation, IndoBERT maintained 96% accuracy with balanced metric distribution, while DeBERTa recorded 93% accuracy. IndoBERT performed better in recognizing neutral sentiment, whereas DeBERTa was more time-efficient. These results highlight the effectiveness of local transformer models and data balancing techniques in improving sentiment classification performance on imbalanced datasets.