🧬 Horizon — Health Intelligence Platform

A unified multi-track ML dashboard for drug toxicity, antibiotic resistance, and epidemic forecasting

---

✨ Overview

Horizon is a Streamlit-based health intelligence dashboard that bundles three independent machine learning tracks into one deployable repository. Each track is fully self-contained — its own model, dataset, artifacts, and UI — all unified under a single navigation hub.

Built for the CodeCure AI Hackathon.

---

🧪 Tracks

Track A — Drug Toxicity Prediction

Molecular toxicity classification across 12 Tox21 endpoints using a tabular XGBoost pipeline with SHAP explainability.

• Model: XGBoost ensemble
• Mean AUC: 0.8514 across 12 endpoints
• Explainability: SHAP feature importance

Track B — Antibiotic Resistance Prediction

Predicts antibiotic resistance from clinical microbiology data using a stacked ensemble with SMOTE-NC balancing.

• Model: Stacked ensemble (XGBoost + LightGBM + CatBoost + Logistic Regression meta-learner)
• AUC: 0.8540 (Optuna-tuned)
• Features: Species resistance rate, antibiotic class resistance rate, interaction terms
• Explainability: SHAP + AI Clinical Advisor (LLaMA 3.3 70B via OpenRouter)

Track C — Epidemic Spread Forecasting

Time-series forecasting and classification of epidemic spread using JHU and OWID data.

• Model: Forecaster + Classifier pipeline
• Data: JHU CSSE + Our World in Data merge

---

🏗️ Repository Structure

Horizon/
├── horizon.py              # Unified Streamlit navigation hub (main entrypoint)
├── requirements.txt        # Repository-wide dependencies
├── runtime.txt             # Streamlit Cloud runtime pin (python-3.12)
├── Dockerfile              # Optional container entrypoint
├── PROJECT_STRUCTURE.md    # Quick file map
├── README.md
├── LICENSE
│
├── track_a/                # Drug Toxicity
│   ├── app.py
│   ├── track_a_pipeline.py
│   ├── data/
│   └── artifacts/
│
├── track_b/                # Antibiotic Resistance
│   ├── app.py
│   ├── track_b_model.py
│   ├── track_b_data_loader.py
│   ├── data/
│   └── artifacts/
│
├── track_c/                # Epidemic Forecasting
│   ├── app.py
│   ├── track_c_model.py
│   ├── track_c_data_loader.py
│   ├── data/
│   └── artifacts/
│
└── docs/                   # Handoff reports and presentation assets

---

🚀 Quick Start

Install dependencies

pip install -r requirements.txt

Run the unified dashboard

streamlit run horizon.py

Run a single track directly

streamlit run track_a/app.py
streamlit run track_b/app.py
streamlit run track_c/app.py

---

☁️ Streamlit Cloud Deployment

Setting	Value
Repository	gaurav-3821/Horizon
Branch	main
Main file path	horizon.py
Python runtime	python-3.12 (from runtime.txt)

---

🐳 Docker (Optional)

docker build -t horizon .
docker run -p 8501:8501 horizon

---

🛠️ Tech Stack

Layer	Tools
UI	Streamlit, custom neobrutalist CSS
ML Models	XGBoost, LightGBM, CatBoost, scikit-learn
Explainability	SHAP
Balancing	SMOTE-NC
Tuning	Optuna
AI Advisor	LLaMA 3.3 70B via OpenRouter
Data	Tox21, Mendeley AMR, JHU CSSE, OWID

---

📏 Design Rules

• One authoritative entrypoint: horizon.py
• One authoritative dependency file: requirements.txt
• Track-local assets stay inside their matching track folder
• No duplicate master-app files
• All paths resolved with pathlib relative to each app's directory

---

👤 Author

Gaurav · @gaurav-3821

---

⭐ Star this repo if you find it useful!