Performance of Machine Learning Algorithms on Imbalanced Sentiment Datasets Without Balancing Techniques

This study explores the performance of five sentiment classification algorithms—Naïve Bayes, Logistic Regression, Support Vector Machine, Decision Tree, and Random Forest—on an imbalanced sentiment dataset, with the SMOTE technique applied as a comparison. The research follows the Knowledge Discovery in Databases (KDD) framework, which includes data selection, preprocessing, transformation, data mining, and evaluation. The evaluation uses metrics such as accuracy, precision, recall, F1-score, and macro average F1-score. Initial results show that all five algorithms performed fairly well even without using a balancing technique, with Naïve Bayes achieving the highest F1-score of 0.84 and recall of 0.81. After applying SMOTE, only small improvements were observed in some models, such as Random Forest (F1-score increased from 0.81 to 0.85), while other models like Naïve Bayes experienced a decrease in performance, dropping to 0.77. This suggests that the effect of balancing techniques like SMOTE can vary depending on the algorithm. Thus, this study provides empirical contributions that highlight the importance of selecting appropriate approaches and the need for a deep understanding of each algorithm's behavior in the context of imbalanced data. Researchers are encouraged to carefully consider these aspects when designing experiments and interpreting results.