<![CDATA[Hybrid learning has become an integral component of higher education, making student feedback an important source of information for evaluating learning quality and improving instructional practices. Although sentiment analysis has been widely applied in educational research, limited attention has been given to how different training–testing data-splitting ratios influence classification performance. This study aimed to compare the performance of the Naïve Bayes algorithm using three data-splitting ratios (60:40, 70:30, and 80:20) for classifying students' sentiments toward hybrid learning at UIN Raden Intan Lampung during the second semester of the 2024–2025 academic year. A quantitative approach was employed to analyze 4,728 student opinions collected from 394 respondents. The analytical procedure consisted of text preprocessing, sentiment labeling using the InSet Lexicon, TF-IDF feature extraction, and Naïve Bayes classification, followed by performance evaluation using accuracy, precision, recall, and F1-score. The findings revealed that 70.3% of the opinions expressed positive sentiment, while 29.7% were negative, indicating that students generally perceived hybrid learning favorably despite challenges related to instructional delivery and internet connectivity. The comparison of data partitions showed similar overall performance, with accuracy values of 79% for the 60:40 split, 78% for the 70:30 split, and 79% for the 80:20 split. However, the 80:20 configuration achieved the strongest class-level performance, producing an F1-score of 87% for positive sentiment and 51% for negative sentiment. These findings demonstrate that Naïve Bayes provides stable performance across different data partitions, while the 80:20 ratio offers the most balanced configuration for sentiment classification within the scope of this dataset and analytical framework.]]>
Copyrights © 2026
