Comparison of Classification Algorithm In Sentiment Analysis of E-Commerce App Comments Based on Feature Extraction and Imbalanced Handling Technique

The rapid growth of e-commerce has significantly changed consumer habits, offering previously unheard-of accessibility and convenience. A massive amount of user-generated content, such as reviews and comments, was produced. Sentiment analysis has advanced, but several problems remain. One major issue was the imbalance in user reviews, where a positive comment is more than a negative comment. The proposed goal was to compare the performance of a classification algorithm that applied feature extraction and overs amp ling to the sentiment analysis of comments written on e-commerce apps. The dataset was taken from Google Plays about e-commerce, with 1500 data in Indonesian comments and ratings. Based on the results, both the Overs amp ling (SMOTE) and undersampling techniques as imbalanced handling methods can enhance the test accuracy of all algorithms like Artificial Neural Networks (ANN), Naive Bayes (NB), Random Forests (RF), and Support Vector Machines (SVM) models using TF-IDF as feature extraction. The results suggest that Rus outperformed SMOTE in ANN, NB, and SVM accuracy by around 85%, 85%, and 84.6%, respectively. In contrast, Smote boosted the RF model by 80.8%. Overall, The NB model has boosted accuracy testing by roughly 8%, while imbalance data (76.67%) has grown to 85%.