<![CDATA[Learning styles play an important role in supporting students' academic achievement, because a mismatch between learning methods and student learning preferences can reduce learning motivation, hinder material comprehension, and have a negative impact on academic results. To address this issue, a web-based learning style prediction system was developed in this study by applying the Random Forest approach. However, in classification problems, Random Forest often faces suboptimal classification results. This study seeks to enhance the effectiveness of the Random Forest algorithm by applying hyperparameter tuning techniques, specifically through grid search and random search methods, in order to improve the accuracy of classifying students' learning styles. This study uses a secondary questionnaire dataset consisting of 1,210 data samples classified based on the VAK model of learning preferences (Visual, Auditory, and Kinesthetic). The test results show that the model with hyperparameter optimization using random search can improve accuracy performance by 88% and the average cross-validation score by 86%. Compared to the default configuration model and grid search, which only achieved an accuracy of 86%, this improvement was also demonstrated by data balancing using SMOTE and data reduction using PCA, which contributed to enhancing model accuracy. Overall, this study demonstrates that grid search and random search hyperparameter optimization can optimize the Random Forest algorithm for accurately classifying students' learning styles. ]]>
Copyrights © 2025
