A Novel Ensemble Feature Selector and Stacked Learning Models for an Dance Display Systems

The integration of advanced machine learning techniques in performing arts has gained increasing attention, particularly in designing intelligent dance display systems that can automatically recognize and classify dance movements with high precision. The traditional machine learning methods often suffer from limitations, including over fitting, reduced generalization ability, and sensitivity to noise, which hinder their effectiveness in complex and dynamic datasets. To address these drawbacks, this study proposes a hybrid feature selectors and stacked classification framework that combines K-Nearest Neighbourhood (KNN), Random Forest (RF), Support Vector Machine (SVM), leveraging the ensemble feature selection which consists of Pearson Coefficients (PC) and Boruta Models. The model was trained and evaluated using the Kaggle Dance Movement Datasets, which consists of labelled sequences of diverse dance postures. The proposed RF-SVM model experimental results attains an overall accuracy of 0.99, with a precision of 0.982, recall of 0.98, specificity of 0.98%, and an F1-score of 0.98%, significantly outperforming other conventional algorithms. Comparative analysis confirms that the hybrid approach enhances recognition performance, reduces misclassification, and ensures greater robustness in real-world applications. This makes the system highly applicable for smart dance training, choreography assistance, and interactive performance platforms.