Stroke Prediction Using Deep Learning

A state-of-the-art deep learning project for predicting stroke risk based on patient health data.

Problem Statement

Stroke is a leading cause of death and disability worldwide. Early detection and risk assessment can significantly improve patient outcomes. This project aims to develop a reliable deep learning model for predicting stroke risk based on patient demographic and health data.

Project Structure

stroke-prediction/
├── data/                 # Data storage and processing
│   ├── raw/              # Raw dataset files
│   └── processed/        # Processed dataset files
├── models/               # Saved model checkpoints
├── notebooks/            # Jupyter notebooks for exploration and visualization
├── src/                  # Source code
│   ├── data/             # Data processing utilities
│   ├── models/           # Model architecture definitions
│   ├── training/         # Training scripts
│   ├── evaluation/       # Evaluation scripts
│   └── utils/            # Helper functions
├── tests/                # Unit tests
├── app/                  # Web application for deployment
├── requirements.txt      # Python dependencies
└── README.md             # Project documentation

Installation

1. Clone this repository:

git clone https://github.com/yourusername/stroke-prediction.git
cd stroke-prediction

2. Create a virtual environment (optional but recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

Dataset

This project uses the Stroke Prediction Dataset from Kaggle, which is publicly available and contains the following features:

• Demographic information (age, gender)
• Medical history (hypertension, heart disease)
• Lifestyle factors (smoking status, work type)
• Health metrics (BMI, glucose level)
• Target: Stroke occurrence (0 = No, 1 = Yes)

Usage

Data Preprocessing

python src/data/preprocess.py

Model Training

python src/training/train.py

Model Evaluation

python src/evaluation/evaluate.py

Web Application

python app/app.py

Then open your browser and navigate to http://localhost:5000

Model Architecture

The project employs a multi-layer neural network with specialized handling for categorical and numerical features. Techniques include:

• Feature normalization and encoding
• Embedding layers for categorical variables
• Dropout for regularization
• Batch normalization
• Class imbalance handling with weighted loss

Performance

The model achieves:

• AUC-ROC: ~0.85
• F1-Score: ~0.78
• Precision: ~0.76
• Recall: ~0.81

Compute Resources

This project is designed to run on free compute resources:

• Training: Google Colab (with free GPU runtime)
• Inference: Local CPU or Colab
• Deployment: Lightweight enough for Heroku free tier

License

This project is licensed under the MIT License - see the LICENSE file for details.