In recent years, the intersection of computer vision and yoga practice has emerged as a promising area of research, focusing on developing automated systems for accurate recognition and analysis of yoga postures. Action recognition can benefit immensely from shape descriptors called invariant moments since they are unaffected by rotation, translation, and scaling variations. This work aims to thoroughly evaluate distinct machine-learning techniques in the context of posture recognition, using invariant moments as key feature descriptors. Three different types of invariant moments, namely raw moments, central moments, and hu moments, were applied as input features for the machine learning models. This research is focused on classifying five distinctive yoga poses: Camel, Cat-Cow, Downdog, Lotus, and Wheel. The Support Vector Machine exhibited the best performance when combined with hu-moments and attained the highest accuracy rate of 96.30% among the many combinations of invariant moments and machine learning algorithms examined.