Clinical Triage and Prediction API

This repository presents the architecture of a predictive system for clinical decision support, ready for cloud deployment.

Architecture and Data Pipeline

Covers data extraction, machine learning model training, and containerized deployment.

Tech Stack: Python, SQL, Docker, FastAPI, Uvicorn, Pydantic, Scikit-learn, Pandas, Joblib.

Pipeline and Methods:

• Data Engineering & Deployment: Relational data extraction via SQL (Mock Data) and production via containerized REST API (Docker) with strict type validation.
• Recommendation System: Instance-based learning (Cosine Similarity) for identifying analogous clinical cases.
• Supervised Models: Random Forest, Extra Trees, Multilayer Perceptron (MLP).
• Unsupervised Models: KMeans, GaussianMixture, SpectralClustering, DBSCAN, Birch, MiniBatchKMeans, AgglomerativeClustering, OPTICS, MeanShift.

Evaluation Metrics:

• Classification: Confusion Matrix, Accuracy, Precision, Recall/Sensitivity, F1-Score, ROC Curve (AUC).
• Clustering: Silhouette Score, Davies-Bouldin, Calinski-Harabasz, Adjusted Rand Index.

The clinical data used consists of publicly available mean values and does not contain private or sensitive patient records.