🩺 AI-Powered Healthcare System

An end-to-end ML/AI project that predicts stroke risk, clusters patients into meaningful groups, and generates treatment recommendations using association rule mining. It also includes a Streamlit app for interactive predictions.

---

📌 Features

• Data Preprocessing

  • Cleans and encodes the stroke dataset
  • Handles missing values and categorical encoding
  • Saves processed data

• **Supervised Learning **

  • Logistic Regression, Random Forest, XGBoost baselines
  • 5-fold Cross Validation
  • Evaluation with ROC AUC, Precision, Recall, F1
  • SHAP-based feature importance

• **Unsupervised Learning **

  • KMeans & DBSCAN clustering
  • Cluster profiles with mean feature summaries
  • Risk-based cluster naming (High/Moderate/Low Risk groups)

• Association Rules

  • Simulated patient symptoms → treatments transactions
  • Apriori + FP-Growth mining
  • Top-10 rules exported for recommendations

• **Streamlit App **

  • Single patient risk prediction
  • Cluster assignment with profile interpretation
  • Recommended treatments from association rules

---

🏗️ Project Structure

ai-healthcare-system/
│
├── data/
│ ├── raw/ # Original dataset
│ ├── processed/ # Cleaned & preprocessed data
│ │ └── stroke_data_processed.csv
│
├── models/ # Trained & saved models
│ ├── model.pkl # Best supervised model (LogReg / RF / XGB)
│ ├── kmeans.pkl # Saved KMeans clustering model
│ ├── scaler.pkl # Scaler used for clustering
│
├── notebooks/ # Jupyter notebooks (experiments & reports)
│ ├── 01-eda.ipynb # Exploratory Data Analysis
│ ├── 02-supervised-baseline.ipynb# Baseline supervised models
│ ├── 03-clustering.ipynb # Clustering experiments
│ ├── 04-association.ipynb # Association rule mining
│
├── src/ # Source code
│ ├── data/
│ │ ├── load.py # Load raw/processed data
│ │ └── preprocess.py # Data cleaning & feature engineering
│ │
│ ├── models/
│ │ ├── baseline.py # Pipelines for baseline models
│ │ ├── trainer.py # Training & cross-validation
│ │ └── evaluate.py # Model evaluation & metrics
│ │
│ ├── unsupervised/
│ │ └── clustering.py # KMeans & DBSCAN + cluster profiling
│ │
│ ├── association/
│ │ └── apriori_rules.py # Association rules (Apriori/FP-Growth)
│ │
│ └── app/
│ └── streamlit_app.py # Streamlit web app integration
│
├── cluster_profiles.md # Cluster summaries (generated in Milestone 3)
├── association_rules.csv # Top-10 rules (generated in Milestone 4)
├── requirements.txt # Python dependencies
├── README.md # Project documentation
└── .gitignore # Files ignored by Git

⚙️ Installation & Setup

1. Clone the repo:

   git clone https://github.com/yourusername/ai-healthcare-system.git
   cd ai-healthcare-system

2.Create a virtual environment:

conda create -n ai-healthcare python=3.10 -y
conda activate ai-healthcare

3.Install dependencies:

  pip install -r requirements.txt

⚡ Execution Workflow

Follow this order to execute the project end-to-end:

---

# Run data loading
python src/data/load.py

# Run preprocessing
python src/data/preprocess.py

# Train baseline models (LogReg, RF, XGBoost)
python src/models/trainer.py

# Evaluate saved model
python src/models/evaluate.py

# Run clustering (KMeans + DBSCAN)
python src/unsupervised/clustering.py

# Mine Apriori + FP-Growth rules
python src/association/apriori_rules.py

# Launch the interactive app
streamlit run src/app/streamlit_app.py

📈 Example Outputs 🧠 Model Performance

Logistic Regression (best CV ROC AUC ≈ 0.84)

XGBoost: tunable for higher recall/precision

🌀 Clustering

KMeans Silhouette Score ≈ 0.15

Clusters:

Cluster 0 → High-Risk Group

Cluster 1 → Moderate-Risk Group

Cluster 2 → Low-Risk Younger Group

📋 Example Rule

symptom:hypertension, symptom:obese → treatment:antihypertensive, treatment:lifestyle_change
(Lift: 18.51, Confidence: 1.00)

📸 Screenshots

🏠 Home Page

🧑‍⚕️ Single Patient Prediction

🛠️ Tech Stack

Programming Language

• Python 3.10+

Libraries & Frameworks

• Data Handling: pandas, numpy

• Visualization: matplotlib, seaborn

• Machine Learning: scikit-learn, xgboost, imblearn

• Clustering: scikit-learn (KMeans, DBSCAN)

• Association Rules: mlxtend (Apriori, FP-Growth)

• Explainability: shap

• App Framework: streamlit

• Serialization: joblib

  🚀 Future Improvements

• Deploy via Docker or Cloud: Package the app using Docker or deploy on platforms like Heroku, AWS, or GCP for wider accessibility.

• Integrate Real Clinical Datasets: Incorporate real-world patient datasets with treatment + outcome mappings to improve the reliability of predictions.

• Temporal Association Rules: Enhance the association rule mining by including temporal patient history (sequence of symptoms → treatments → outcomes).

• Improved Interpretability: Add interactive LIME/SHAP dashboards within the app for doctors and researchers to better understand model decisions.