ECZEMA DETECTION PROJECT

A simple machine learning-powered web app to detect eczema from skin images.

📁 Dataset

Download here: Google Drive Dataset Link
Contains labeled skin images organized into:

• eczema/
• normal/

How It Works

1. We trained a Support Vector Machine (SVM) model using labeled images of eczema and normal skin.
2. The model is deployed in a Streamlit web app for interactive image-based prediction.
3. Users upload a skin image.
4. The app preprocesses the image and predicts whether it's eczema or normal.
5. The result is displayed clearly on screen.

---

Project Structure

eczema_project/
│
├── dataset/              # Training data (eczema / normal images)
│   ├── eczema/
│   └── normal/
│
├── model_train.py        # Trains multiple ML models
├── eczema_model.pkl      # Saved trained model (SVM)
├── app.py                # Streamlit app for predictions
├── requirements.txt      # Project dependencies
└── .gitignore            # Version control exclusions

Models Used

We trained and tested multiple models to identify the best performer:

Model	Description
SVM	Best performance (~79.3% accuracy), used in deployed app
Random Forest	Robust ensemble model, good with imbalanced data
K-Nearest Neighbors	Simple baseline model, less effective for image data
MLP (Neural Net)	Basic neural network with 2 hidden layers, decent performance
CNN (Deep Learning)	Custom CNN model trained with TensorFlow; limited due to hardware constraints

Only SVM is used in the current version of the app, but other models are available for experimentation.

---

How to Run the Project

1. Clone the repository

git clone <your-repository-url>
cd eczema_project

2. (Optional) Create a virtual environment

python -m venv venv
# On Windows
.\venv\Scripts\activate
# On Mac/Linux
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. (Optional) Train the model from scratch

python model_train.py

This will retrain all models and generate eczema_model.pkl.

5. Run the Streamlit web app

streamlit run app.py

6. Use the App Open the URL (e.g. http://localhost:8501) in your browser and:

• Upload a skin image
• Get instant prediction
• Try with different images for testing

---

✅ Features

• 📷 Upload skin images to classify as eczema or normal
• ⚙️ Trained on real image data
• 💾 Multiple models tested, best selected
• 🌐 Simple and responsive Streamlit web interface

---

🔮 Future Improvements

• 🧠 Train deeper CNN models for better accuracy
• 📈 Add support for real-time camera input
• 🧬 Collect more diverse and larger dataset
• 🧪 Test with pre-trained models (e.g., ResNet, MobileNet)
• 🎨 Improve UI/UX of the web interface

---

🛠️ Tech Stack

• Python
• Scikit-learn
• Streamlit
• NumPy
• Pillow (PIL)
• TensorFlow (for CNN training)

---

📂 Requirements

Make sure you have the following installed (or use the provided requirements.txt):

numpy
pillow
scikit-learn
streamlit
joblib
tensorflow
imblearn
matplotlib

---

💡 Credits

Supervisor : Dr. Amit Kumar

Created by a team of students exploring AI in medical imaging with limited resources — proving that even low-end hardware can still deliver smart solutions