POSTURE DETECTION USING XCEPTION AND SVM WITH EXPLAINABLE AI (SHAP).

ABSTRACT This project presents a hybrid human posture detection system that integrates deep learning and machine learning techniques for accurate and explainable classification of human postures. The proposed approach combines the Xception model for deep feature extraction with a Support Vector Machine (SVM) classifier for final posture categorization. The Xception model leverages depthwise separable convolutions to efficiently capture spatial and structural characteristics from posture images, while the SVM ensures robust decision boundaries for multi-class classification. To enhance transparency, SHAP (SHapley Additive exPlanations) is employed to interpret model predictions, offering insights into how specific features influence classification outcomes. The system was evaluated using a dataset of yoga postures, achieving a classification accuracy of 94.01%, precision of 0.89, recall of 0.98, and F1-score of 0.94. The results demonstrate that the hybrid Xception–SVM model achieves a strong balance between computational efficiency, accuracy, and interpretability, making it suitable for applications in healthcare, ergonomic monitoring, and rehabilitation systems where both performance and explainability are critical.