BodyMotionClassification

It is a simple body motion classification implementation. There is a small dataset for training for five actions of 'Walking', 'Running', 'Jumping', 'Punching', and 'Kicking'. After preprocessing, joint-based feature extraction takes place. Extracted features are 'rotations', 'velocity', 'acceleration', 'range_of_motion', 'average_rotation', 'spatial_path', 'harmonics', 'symmetry', 'frequency_analysis', 'joint_distance', 'angular_velocity', 'angular_acceleration'. Finally, the XGBoost trained model will be tested on new unseen data and return a confusion matrix and classification report for final performance evaluation. Also, there are some plots for the test phase as follows: "Violin Plot", "ROC Curve", "Precision and Recall Plot", "Confusion Matrix Heatmap", and "Cumulative Gain Curve". Furthermore, there are metrics for comparing the original and the synthetic samples namely, "Diversity", Dynamic Time Warping "DTW", and Mean Per-Joint Position Error "MPJPE".