Yoga pose detection and classification have gained significant interest due to their applications in fitness, health, and rehabilitation. Existing methods often rely on a large number of keypoints, which can lead to high computational costs and complexity. This work introduces a novel Pose-Critical Keypoint Weighting (PCKW) model designed to address these challenges. By reducing the number of keypoints from 33 to 18, the model enhances computational efficiency while maintaining high classification performance. A new dataset was developed, encompassing nine distinct yoga poses performed by diverse individuals, with keyframes extracted at a rate of 25 frames per second. Evaluation of the PCKW model demonstrated an accuracy of 92.5%, precision of 91.0%, recall of 93.2%, and an F1-score of 92.1%. These results highlight the model’s superior performance and lower computational complexity compared to traditional CNNs, RNNs, SVMs with keypoints, and decision trees, underscoring its effectiveness for real-time yoga pose analysis.