Indian-Population-Pose-Asymmetry-Classifier-v1

About Model

The Indian Population Pose Asymmetry Classifier (IPPAD-PAC v1.0) is a machine learning model trained to classify human postural asymmetry into three categories: Symmetric, Mild Asymmetry, and Severe Asymmetry. The model takes 22 bilateral biomechanical features as input, derived from COCO-style 17-keypoint skeleton data, and outputs a class label along with per-class probabilities and a robot trajectory recommendation. The architecture is a soft-voting ensemble combining a Random Forest (200 trees, max depth 12, balanced class weights) and a Gradient Boosting Machine (150 estimators, learning rate 0.08), with input features standardized via StandardScaler. The model was trained and evaluated on IPPAD v1.0, a 3,000-sample dataset calibrated to Indian anthropometric norms from the SIZE INDIA reference survey. It achieves 90% test accuracy and a macro F1 score of 0.90 across all three classes, with 5-fold cross-validation F1 of 0.916. The model artifact includes trained weights in joblib format, a ready-to-run inference script with robot trajectory advice output, evaluation plots, and full documentation.