How eaglepython/healthcare-analytics-portfolio works
Project type and purpose
• This is a multi-project analytics portfolio repository (“Healthcare-Analytics-Portfolio”) that bundles two production-grade Streamlit applications (Project A: cardiovascular risk prediction; Project B: heart-failure readmission prevention), plus two supplemental dashboards (financial performance and patient outcomes).
• Each sub-project implements an end-to-end ML pipeline (data ingestion/engineering, model training, model storage) and an interactive dashboard UI, all containerized via Docker Compose for local deployment.
Top-level view of the file structure
• Root: demo.py, requirements.txt, Dockerfile, docker-compose.yml
• Two main subfolders under Healthcare-Analytics-Portfolio/:
• cardiovascular-risk-prediction/
• heart-failure-readmission-prevention/
• Two supplemental “Suplemental Projects/” folders: Project-1-Financial-Performance, Project-2-Patient-Outcomes
• Notebook: Real_Time_Warranty_Claim_Anomaly_Detection_System.ipynb
• Each sub-project contains:
• data_generation.py (synthetic or ingestion scripts)
• analysis_pipeline.py or intervention_analysis.py (model training & evaluation)
• dashboard.py or hf_analytics_platform.py (Streamlit UI)
• models/ (serialized ML models)
• reports/ (executive summaries)
Key README insights
• Two production dashboards exposed on different ports via Docker Compose (8501, 8502)
• MLflow service on port 5000 for experiment tracking (defined in docker-compose.yml)
• Data sources: synthetic datasets, EMR API integrations (HL7 FHIR), real-time streaming alerts
• Technology stack: Python 3.8+, Streamlit, XGBoost, Random Forest, Docker, Docker Compose, MLflow
Main components and their interactions
A. Data Sources & Ingestion
• Synthetic data generator (data_generation.py) or real EMR‐FHIR API
• Raw files (.csv/.xlsx) under data/ and ingestion scripts
B. ETL & Feature Engineering
• analysis_pipeline.py / intervention_analysis.py
• Preprocessing → feature store (in memory or local filesystem)
C. Model Training & Model Store
• Multiple algorithms (Logistic Regression, Random Forest, XGBoost, Neural Net, Survival Analysis)
• Outputs saved to models/ directory or tracked in MLflow
D. Dashboard UI (Streamlit)
• dashboard.py / hf_analytics_platform.py expose:
• Real‐time risk calculator
• Population analytics and cohort trends
• Executive KPI dashboards (ROI, cost‐effectiveness)
E. MLOps & Monitoring
• MLflow service for experiment tracking and model registry
• Logging & audit trails for compliance (HIPAA)
F. Container Orchestration
• Dockerfile for each service (two Streamlit apps + MLflow)
• docker-compose.yml wires up:
• cardiovascular risk app (port 8501)
• heart-failure app (port 8502)
• MLflow server (port 5000)
Architectural patterns & design principles
• ETL pipeline (data → features → models → dashboards)
• MVC-like separation:
• Model: analysis_pipeline.py & models/
• View: Streamlit dashboards
• Controller: Streamlit app orchestrates data fetch → model inference → rendering
• Containerized micro-services (each dashboard + MLflow)
• MLOps best practices: experiment tracking, model registry
• Secure by design: role‐based access, encryption, audit logs