UmarMubiru/UGANDA-SIGN-LANGUAGE-INSTRUCTOR-SW-ML-9

Uganda Sign Language Instructor An end-to-end machine learning project for Uganda Sign Language (USL) recognition, focused on disease-related signs, model training, deployment, and interactive demo delivery. This repository combines: - dataset analysis and feature engineering, - supervised model training and selection, - FastAPI deployment for inference, - Docker and PowerShell operational workflows, - Streamlit interface for practical usage. Project Scope The project is organized around two complementary goals: 1. Disease-sign classification pipeline (Dataset 1 style workflow) - Build tabular features from sign language videos. - Train and compare multiple classifiers. - Export the best model artifact for deployment. - Serve predictions through a production-ready API. 2. Interactive USL demonstration workflow (Streamlit) - Support image, video, and webcam-style input paths. - Run lightweight feature extraction for inference. - Provide top-k predictions and confidence outputs. System Architecture ```mermaid flowchart TD A[SIGN LANGUAGE DISEASES FINISHED videos] --> B[Feature Engineering\nprepare_csv_features.py] B --> C[dataset1_disease_features.csv] B --> D[dataset1_disease_splits.csv] C --> E[Train RF + Logistic\ntrain_rf_logreg_disease.py] C --> F[Train SVM + HGB\ntrain_svm_hgb_disease.py] D --> E D --> F E --> G[model_metrics.json] F --> H[model_metrics_svm_hgb.json] G --> I[Best Model Selection\nexport_best_model.py] H --> I I --> J[best_model.joblib] J --> K[FastAPI Inference\nmodels_dataset1/deployment/api.py] J --> L[Streamlit App\nstreamlit_app/app.py] K --> M[REST Clients] L --> N[End Users] ``` Model Benchmarks Benchmarks below are sourced from the repository artifacts: - `models_dataset1/csv_models/artifacts/model_metrics.json` - `models_dataset1/csv_models/artifacts/model_metrics_svm_hgb.json` | Model | Validation Accuracy | Validation Macro-F1 | Test Accuracy | Test Macro-F1 | | -------------------- | ---------- …