An Explainable Machine Learning (XAI) framework for classification of intricate dancing posture among Indian Bharatanatyam dancers
K. Adalarasu, RM. Kuppan Chetty, K. Ghousiya Begum, S. Harini, Mukund Nilakantan Janardhanan
Abstract
India, the cradle of ancient civilizations such as the Indus Valley, has a rich tradition of classical dance forms like Kathak, Bharatanatyam, Kuchipudi, Manipuri, and Odissi. Indian classical dance (ICD), deeply embedded in the cultural fabric, has developed detailed grammar, rules, and stagecraft. Bharatanatyam, one of the oldest and most esteemed dance forms, is known for its coordinated movements, balanced postures, and expressive gestures. Understanding the underlying semantics of performing arts like Bharatanatyam is challenging, especially since long dresses occlude the legs and camera viewpoints affect classification performance. While existing methods achieve varying accuracies in posture recognition, real-world applications often see reduced performance due to these challenges. This study aims to enhance the analysis of Bharatanatyam postures using a force platform to record real-time vertical ground reaction force (VGRF) data. Key contributions include recording real-time force data for six intricate Bharatanatyam postures, extracting VGRF features, classifying postures using machine learning techniques, and implementing Explainable AI (XAI) methods to elucidate feature contributions. The novelty of this approach lies in using force platform data instead of video images, achieving 98.2 % accuracy, and addressing issues like occlusions and background clutter. The comparative results demonstrate that XAI methods outperform traditional classifiers, providing robust solutions for accurate dance posture analysis and reducing expert-based errors in classifications, thereby bridging traditional dance art with cutting-edge technologies. • Novel framework for classifying intricate dance postures using VGRF data. • Six complex dance postures analyzed using ML techniques with accuracy of 98.2 %. • Explainable AI provided insights on classification of the Indian classical dance. • This study bridges traditional Indian dance forms and modern technology. • This study opens avenue for interdisciplinary research in AI, biomechanics, and motion analysis.