AI-DRIVEN YOGA RECOMMENDATION SYSTEM USING DEEP LEARNING FOR REAL-TIME POSE ACCURACY ASSESSMENT

The incidence of lifestyle diseases is increasing at an alarming rate, and yoga has been found to be an effective, evidence-based, non-pharmacological solution for many of these diseases. In this study, an “AI-Driven Yoga Recommendation System” is presented, which consists of three main components: a rule-based system that maps 24 different healthcare conditions to yoga practices, a VGG16 Transfer Learning CNN for four- class pose classification with a 52-image custom dataset, and a MediaPipe BlazePose module that detects 33 real-time skeletal keypoints and offers live feedback. On a validation set of nine samples, the system showed 88.9% accuracy, a macro F1-score of 90.0%, and an AUC score of 1.00 for all four classes. The entire application can be run using a standard webcam and Flask web interface, with no additional hardware requirements. This demonstrates the effectiveness of the Transfer Learning approach with small datasets for real-world yoga coaching systems.