<![CDATA[Personalized learning faces challenges when Information Systems students must choose a study path among many specialization options, while existing systems often fail to map student interests accurately. Static preference data are commonly treated as independent features, which prevents models from capturing relationships between interest scores. This study proposes a solution using a Simple Recurrent Neural Network that represents seven interest scores as a single sequence to capture positional context across features. A dataset of 318 respondents was used for training, and SMOTE was applied to address label imbalance. The model was compared with a Dense Neural Network to evaluate the impact of the sequential representation. SimpleRNN achieved an accuracy of 90.10 percent at 100 epochs, outperforming the DNN result of 80.20 percent. Evaluation using the confusion matrix along with precision, recall, and F1-score showed that SimpleRNN offers more stable classification, especially for interest categories with similar characteristics. These results indicate that applying a sequential approach to static data improves interest classification performance and supports more accurate personalized learning path recommendations.]]>
Copyrights © 2026
