Melanoma Detection using Transfer Learning (ISIC 2020)

A deep learning-based skin cancer classification project using transfer learning to distinguish malignant melanoma from benign skin lesions. Trained and evaluated on the ISIC 2020 Challenge Dataset, this project benchmarks the performance of multiple CNN architectures with metrics across training, validation, and test sets.

---

Dataset

• Source: ISIC 2020: Skin Lesion Analysis Towards Melanoma Detection
• Size: ~45GB dermoscopic image dataset
• Format: JPEG images with binary labels (benign vs malignant)

---

Models Used

The following pre-trained models were fine-tuned on the dataset:

• MobileNetV2
• EfficientNetB0
• DenseNet121
• ResNet50
• VGG16

All models were trained for 20 epochs with:

• Image size: 224×224
• Optimizer: Adam
• Loss Function: Binary Crossentropy
• Augmentations: Rotation, flipping, brightness, and zoom
• Class balancing: Downsampling majority class

---

Performance Metrics

Model	Train Accuracy	Val Accuracy	Test Accuracy	Train Loss	Val Loss
ResNet50	97.55%	80.91%	77.33%	0.0623	1.5407
MobileNetV2	96.56%	79.93%	82.94%	0.0890	1.0626
DenseNet121	91.23%	77.45%	83.15%	0.2101	0.8533
VGG16	90.07%	78.70%	78.89%	0.2318	0.9274
EfficientNetB0	90.34%	77.24%	80.69%	0.2249	0.8320

Best test accuracy: DenseNet121 – 83.15%
Best training accuracy: ResNet50 – 97.55%

---

Project Structure

├── notebooks/
│   ├── preprocessing.ipynb     # Data loading, cleaning, augmentation
│   ├── training_*.ipynb        # Model training scripts for each architecture
│   └── evaluation.ipynb        # Test set prediction and metrics comparison 
├── requirements.txt            # Python dependencies
└── README.md                   # You're here!

How to Run

1. Clone the Repository

git clone https://github.com/alloy77/Melanoma_Detection.git
cd Melanoma_Detection

2. Download the dependencies

pip install -r requirements.txt

3. Download the ISIC Dataset

Updata paths in notebooks if needed.

4. Run Preprocessing and Training

jupyter notebook notebooks/preprocessing.ipynb
jupyter notebook notebooks/training_mobilenet.ipynb  # or any other model

5. Evaluate the models

jupyter notebook notebooks/evaluation.ipynb

Visualisation

This project includes

• Training and validation accuracy/loss plots
• Confusion matrices on test predictions
• ROC-AUC curves for model comparison
• Class distribution and image augmentation previews

Contributions

Contributions are welcome! If you find a bug or want to propose a new feature/model, feel free to open an issue or submit a pull request. To contribute:

• Fork the repository
• Create a new branch (git checkout -b feature-branch)
• Push to the branch (git push origin feature-branch)
• Open a pull request