<![CDATA[Identifying students’ interests at an early stage is a crucial process in education because it enables schools to direct students’ potential in a structured and objective manner. However, report card data typically contain many attributes, some of which are redundant or irrelevant, and this condition can degrade the performance of classification algorithms such as Naive Bayes. This study compared the classification performance of three models, namely pure Naive Bayes, a hybrid PSO-Naive Bayes model, and a hybrid PCA-PSO-Naive Bayes model, in classifying the interests of elementary school students into four categories: Academic, Arts, Sports, and ICT. The dataset combined cognitive data in the form of report card scores in nine subjects with affective data obtained from an interest questionnaire. Particle Swarm Optimization (PSO) was applied as a feature selection technique, while Principal Component Analysis (PCA) was applied as a dimensionality reduction technique before feature optimization. Model performance was evaluated using stratified 5-fold cross validation. The results showed that the PCA-PSO-Naive Bayes model achieved the highest mean accuracy of 98.92%, compared with 97.84% for pure Naive Bayes and 97.30% for PSO-Naive Bayes. The hybrid PCA-PSO-Naive Bayes model also produced the most stable accuracy across folds, with a range of 97.3% to 100%. These findings indicate that combining PCA-based dimensionality reduction with PSO-based component selection improves both the accuracy and the stability of Naive Bayes in classifying student interests based on academic and questionnaire data.]]>
