<![CDATA[This study investigates the classification of Kodály hand signs using hand landmark features extracted from MediaPipe Hands. A landmark-based approach was chosen because it represents hand gestures as structured numerical features that are more efficient than raw image data. The objective of this study is to compare the performance of three tree-based machine learning algorithms, namely Decision Tree, Random Forest, and XGBoost, for classifying eight Kodály hand sign classes (Do–Do’). A dataset consisting of 8,000 samples was collected with a balanced distribution of 1,000 samples per class from 2 participants using the right hand only. Each sample is represented by 63 features derived from 21 hand landmarks with (x, y, z) coordinates. The data were divided using an 80% training and 20% testing holdout scheme and further validated using Stratified 5-fold cross-validation. Model performance was evaluated using accuracy, macro-precision, macro-recall, macro F1-score, and confusion matrix analysis. Experimental results show that XGBoost achieves the best performance with an accuracy of 0.9712 and macro F1-score of 0.9714 in the holdout evaluation and remains stable in cross-validation with an accuracy of 0.9768 ± 0.0046. Random Forest achieved slightly lower performance, while Decision Tree produced the lowest accuracy. Confusion matrix analysis indicates that most misclassifications occur between classes with similar landmark patterns, particularly La and Mi. Therefore, XGBoost is recommended as the most effective model for landmark-based Kodály hand sign classification. ]]>
Copyrights © 2026
