<![CDATA[The rapid growth of digital applications has heightened the need to understand user perceptions more thoroughly, particularly through sentiment analysis of user-generated reviews. In practice, sentiment classification often faces challenges related to class imbalance, especially when neutral reviews are significantly fewer than positive or negative ones. This imbalance can limit a model’s ability to accurately detect all sentiment categories. This study examines the comparative performance of three machine learning algorithms Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) by applying the Adaptive Synthetic Sampling (ADASYN) technique to address class imbalance. This study differs from previous similar research by conducting a simultaneous comparative analysis of three algorithms using the ADASYN method in the context of Access by KAI application reviews, which has not been examined in prior studies. Experimental results indicate that after implementing ADASYN, model accuracies reached 75.17% for SVM, 84.06% for RF, and 83.17% for XGBoost. Although accuracy slightly decreased after oversampling, the F1-scores for the neutral class improved to 0.13 (SVM), 0.05 (RF), and 0.14 (XGBoost). Before applying ADASYN, the models achieved accuracies of 85.88% (SVM), 85.13% (RF), and 85.37% (XGBoost), but they were unable to effectively recognize neutral sentiments, with F1-scores of 0.00 for SVM and RF, and 0.03 for XGBoost. These findings suggest that ADASYN enhances model sensitivity to neutral sentiment, with XGBoost demonstrating the most consistent and robust performance in sentiment classification for the Access by KAI application.]]>
Copyrights © 2025
