Enhancing Student Sentiment Classification on AI in Education using SMOTE and Naive Bayes

Abstract

This study investigates student sentiment regarding the use of artificial intelligence (AI) in education, employing the Naive Bayes model enhanced with the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance issues. Class imbalance, a common challenge in sentiment classification, often skews model performance toward majority classes, reducing its effectiveness in recognizing minority classes. To mitigate this, SMOTE was applied to generate synthetic samples for minority classes, achieving a more balanced class distribution. The results demonstrate that incorporating SMOTE improved the Naive Bayes model's accuracy from 65% to 78.87% and significantly increased sensitivity to minority classes. Evaluation metrics, including precision, recall, and F1-score, showed satisfactory performance for certain classes, notably classes 2 and 4. However, challenges remained with class 1, where classification accuracy was lower, indicating inherent complexities in its data patterns. While SMOTE successfully enhanced model performance, it also introduced a potential risk of overfitting, particularly with limited original datasets, highlighting the importance of data quality and size. This research offers actionable insights for educators, developers, and policymakers, emphasizing the need for AI systems in education that are adaptive and responsive to student perceptions. The study concludes that Naive Bayes combined with SMOTE is an effective approach for sentiment analysis in imbalanced datasets. Future research should explore more sophisticated models and larger datasets to achieve more comprehensive and representative outcomes.