Yoga Asana Classification Using LightGBM and Swarm Optimization Technique

Yoga has gained global popularity as a means of enhancing physical and mental well-being. Although in-person training offers optimal benefits, it is often inaccessible due to time constraints and financial limitations. This research introduced a novel approach that combines machine learning with advanced optimization techniques. MoveNet has employed to identify essential features from the input data. Subsequently, Principal Component Selection is applied to identify the most significant features, reducing dimensionality and optimizing computational efficiency. Classification is performed utilizing a Light Gradient Boosting Machine (LGBM) to ensure efficiency and accuracy with structured data. The LGBM model has further optimized using the Chicken Swarm Optimization (CSO) algorithm, which fine-tuned hyper-parameters to enhances performance. Experimental outcomes demonstrated the effectiveness of the integrated framework in accurately recognizing yoga poses. The proposed approach achieved a high accuracy with 99 % in yoga pose recognition. The aim of this work is to develop an accessible and affordable system to provide personalized guidance in practicing yoga effectively and safely at home while mitigating the risk of injuries.