A graph neural network-based system for evaluating dance training effectiveness and dynamically adjusting training plans

The integration of digital technologies in dance education provides opportunities to enhance training effectiveness. Traditional motion-based systems primarily focus on static recognition and fail to model complex movement relationships or adapt training plans in real time. This research aims to develop an intelligent framework for evaluating the effectiveness of dance training and dynamically adjusting training plans using a Bobcat optimization-tuned graph neural network with autoencoder (BO-GNN-AE). A Python-implemented system is designed, integrating structured pose keypoints, motion trajectories, and instructor-annotated performance scores. The Dance Heritage Visual Dataset, encompassing eight Indian classical dance forms at low, medium, and high-performance levels, enriches model learning. Data preprocessing employs Kalman filtering for noise reduction to manage missing joints and variable recording speeds. Histogram of oriented gradients (HOG) captures body shape, movement direction, and dynamic posture variations, while local gabor binary pattern – three orthogonal planes (LGBP-TOP) is used for capturing robust spatial–temporal texture features. The Autoencoder compresses high-dimensional descriptors into compact latent features, while Bobcat Optimization tunes GNN-AE hyperparameters to enhance convergence and generalization. The system evaluates rhythm consistency, joint coordination, posture precision, and interaction patterns to generate adaptive practice recommendations. Experimental implementation in Python achieved an F1-score of 0.983, a precision of 0.986, and a recall of 0.982, outperforming conventional neural architectures. The proposed BO-GNN-AE framework enables precise, real-time evaluation and adaptive adjustment of dance training, fostering improved learning outcomes and personalized performance optimization in modern dance education.