🧠 Ischemic Stroke Prediction Using Hybrid Deep Learning Models

This is a final-year academic project aimed at predicting the likelihood of ischemic stroke using a hybrid deep learning model that combines CNN and LSTM architectures. The system integrates a Django-based web frontend for user interaction and RESTful API endpoints for prediction, backed by trained deep learning models.

---

🚀 Objective

To build a robust, accurate, and interpretable prediction model for ischemic stroke using a combination of temporal and spatial health data.

---

📁 Project Structure

📆 stroke-prediction-deep-fusion
🔹 BACKEND/
🔹 └── Ischemic_Stroke.ipynb       # Model training & evaluation notebook (CNN, LSTM, CNN-LSTM)
🔹 FRONTEND/
🔹 ├── self/                       # Django project configuration
🔹 └── webapp/                     # Django app with views, templates, URLs
🔹 DATASET/                        # Stroke dataset CSV
🔹 cnn.pkl                         # Trained CNN model
🔹 lstm.pkl                        # Trained LSTM model
🔹 cnn_lstm.pkl                    # Trained CNN-LSTM hybrid model
🔹 requirements.txt                # Python dependencies
🔹 stroke-prediction-deep-fusion.pdf.pdf  # Full project documentation

---

🧠 Models Used

• CNN: Captures spatial patterns in structured/tabular data.
• LSTM: Models temporal health data dependencies.
• CNN-LSTM: Combines both for improved performance.

All models are trained and saved as .pkl files and can be loaded for predictions via Django views.

---

🌐 Frontend (Django Web App)

• Homepage: index.html – Welcome screen
• Input Page: input.html – Collects user health parameters
• Result Page: output.html – Displays stroke risk prediction

The prediction logic is implemented using a predict() function inside views.py, which loads the appropriate .pkl model and returns the stroke risk prediction.

---

🔧 Backend Processing

• Data is collected from users via forms.
• Data is preprocessed (normalized, encoded).
• The selected model is used to generate a risk prediction.
• Output is rendered on output.html.

---

🛠️ Dependencies

pip install -r requirements.txt

Or manually:

• Django 3.0.8
• pandas, numpy
• scikit-learn
• tensorflow, keras
• matplotlib, seaborn
• djangorestframework
• joblib, Pillow

---

⚙️ How to Run Locally

1. Clone and Setup Virtual Environment

git clone https://github.com/AnandKumar56/stroke-prediction-deep-fusion.git
cd stroke-prediction-deep-fusion
python -m venv venv
source venv/bin/activate   # or venv\Scripts\activate on Windows
pip install -r requirements.txt

2. Run the Django Frontend

cd FRONTEND
python manage.py migrate
python manage.py runserver

Access at: http://127.0.0.1:8000

---

📊 Evaluation Metrics

Model	Training Accuracy	Validation Accuracy
CNN	92.35%	88.76%
LSTM	84.40%	81.92%
CNN-LSTM	96.47%	94.53%

---

👥 Team Members

• K. Goutham – Model Implementation
• K. Arun Kumar – Frontend Developer
• D. Anand Kumar – Backend Developer
• M. Jaswanth – Data Research & Analysis

---

📄 Project Documentation

A detailed write-up of this project is available in the file stroke-prediction-deep-fusion.pdf. It includes the system overview, model architectures, preprocessing steps, project structure, implementation logic, and testing considerations.

---

📄 License

This project is intended for academic use only.

---