🧠 Skin Cancer Classification (ISIC 2024)

This project explores the classification of skin lesions using deep learning and machine learning techniques.
It fuses dermoscopic image data with structured clinical metadata to improve melanoma detection.

---

📦 Project Status

• ✅ EDA on image metadata
• ✅ Preprocessing pipelines for clinical variables
• ✅ Baseline models: Random Forest, XGBoost
• ✅ K-fold stacking (RF, XGB, FFN) + logistic meta-model
• ✅ Initial CNN trial (pretrained backbone)
• ⏳ Hybrid CNN (images + metadata)
• ⏳ Explainability (e.g., SHAP)
• ⏳ Full comparison: image-only vs metadata-only vs hybrid

---

📁 Dataset

The dataset comes from the Kaggle ISIC 2024 Challenge.
Due to licensing and size constraints, raw images are not included.

📄 See data/README.md for download instructions and the expected folder structure.

---

🚀 Reproduce

# 1) Environment
pip install -r requirements.txt

# 2) Data
#   - Follow data/README.md to place CSVs and images under data/

# 3) Stacking (RF + XGB + FFN → Logistic meta)
python -m src.train.stacking_kfold

# 4) CNN trial (pretrained backbone)
python -m src.train.train_cnn

Outputs & Models

• Small artifacts (metrics JSON/CSV, a few PNG figures) will go under outputs/.
• Trained models (.pkl, .keras) are saved under models/.
  Large files are excluded from Git; consider GitHub Releases or external storage and link them here.

---

🧰 Tools & Libraries

• Languages: Python
• Frameworks: TensorFlow/Keras, Scikit-learn, XGBoost
• Visualization: Matplotlib, Seaborn
• Utilities: NumPy, Pandas, UMAP, t-SNE, OpenCV (optional)

---

📂 Project Structure

skin-cancer-classification/
├── data/                 # Raw data (not in repo) + README with instructions
├── notebooks/            # Notebooks as reports (call into src/)
├── src/                  # Source code (pipelines, models, training, eval)
│   ├── config.py
│   ├── data_loader.py
│   ├── preprocessing.py
│   ├── eda_utils.py
│   ├── models/
│   │   ├── cnn.py        # CNN / hybrid CNN builders
│   │   └── tabular.py    # RF, XGB, FFN builders
│   ├── train/
│   │   ├── train_cnn.py
│   │   └── stacking_kfold.py
│   ├── eval/
│   │   ├── metrics.py
│   │   └── plots.py
│   └── utils/
│       ├── seeding.py
│       └── io.py
├── models/               # Trained models (ignored by git; keep README)
├── outputs/              # Curated metrics/figures (small files)
├── references/           # bibliography.md and related notes
├── LICENSE
├── README.md
└── .gitignore

---

🔎 Notes on Reproducibility

• Seeds fixed where applicable (src/utils/seeding.py).
• Configurable parameters in src/config.py.
• If you use or adapt external utilities, keep clear attribution and license notes.

---

🙌 Credits

Some utilities and ideas are inspired by Ilya Novoselskiy — https://github.com/ilyanovo.
See CREDITS.md and respect original licenses when reusing third-party code.

---

✍️ Author

Matteo Morella
📧 Email: mmorella9@gmail.com
🐙 GitHub: @mat126
📄 CV: Download CV – PDF