A machine learning based personalized Yoga Asanas Recommendation Engine

All healthcare therapies are grouped into mainstream (allopathy) and complementary and alternative therapies (CAT). Yoga is a popular therapy that falls under CAT. People practice yoga to maintain and improve their health. Yoga consists of hundreds of postures, each with its own benefits and contraindications. Practicing Yoga requires a careful selection of appropriate asanas (that loosely translates to postures) at an individual level. We investigated the top 100 websites provided by Google when we searched for yoga asanas keywords. We discovered that all these websites provide generic recommendations to the users. Such recommendations could harm rather than help a user. In this work, we propose a personalized Yoga Assistant (YA) that consists of a data collection and preprocessing module and an Asanas Recommendation Engine (ARE). We have implemented ARE using state-of-the-art natural language processing (NLP) techniques, including BERT (Bidirectional Encoder Representations from Transformers), for data encoding. Ours is a multi-label learning model that uses the cosine similarity function to rank classes. We curated and developed a multi-label dataset consisting of 140 asanas that are easy (for beginners) or of intermediate difficulty level, enabling us to conduct thorough investigations into yoga asanas recommendations. The system suggests up to 15 asanas to a user based on her health conditions that she can safely practice. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.