Abstract
In recent years, evaluating competencies such as knowledge, practical skills, character traits, and meta-learning capabilities has gained increasing importance in educational research. As educational datasets grow larger and more complex, machine learning offers promising tools for analyzing student responses and identifying patterns that support assessment processes. This study aims to classify student responses collected through the 21st Century Competencies Scale using a variety of machine learning algorithms, including SVM, ANN, k-NN, RF, LR, DT, AdaBoost, Gradient Boosting, and XGBoost. The dataset contains responses from 616 participants and covers four key sub-dimensions. Model performance was measured using accuracy, precision, recall, and F1-score. Grid search optimization was also applied to improve performance. The highest classification accuracy was achieved by LR in the "Character" sub-dimension (78.73%), followed by SVM in the "Skills" (78.58%) and overall scale (74.51%). Gradient Boosting and k-NN models also showed competitive results across multiple dimensions. These findings emphasize the effectiveness of machine learning, particularly when combined with parameter optimization, in supporting data-driven educational assessments.