This repository contains the implementation of the Skin Lesion Classification project, designed to classify medical images of skin lesions. This project employs an ensemble of LightGBM, CatBoost, and EfficientNet-B0 models, with advanced preprocessing and feature engineering techniques to enhance classification performance.
- Model Ensemble: Combines the strengths of LightGBM, CatBoost, and EfficientNet-B0 for improved accuracy.
- GeM Pooling: Implemented Generalized Mean Pooling for large-scale image datasets.
- Feature Engineering: Engineered 20+ features for the model.
- Advanced Preprocessing: Applied various preprocessing techniques on medical images, including resizing, normalization, and augmentation.
- Skin Lesion Dataset: Publicly available medical image dataset consisting of skin lesion images. Please refer to the official dataset source for download instructions.
- EfficientNet-B0: Pre-trained convolutional neural network fine-tuned for the task of skin lesion classification.
- LightGBM & CatBoost: Gradient boosting models used for tabular feature classification, forming part of the ensemble.
- Accuracy: Achieved high classification accuracy on the validation set.
- Benchmarking: Performance evaluated on key metrics like precision, recall, and F1-score.
To set up the environment and run the project:
git clone https://github.com/heyyviv/Skin-Lesion-Classification.git
cd Skin-Lesion-Classification
pip install -r requirements.txtTo run the project notebooks, you can access them on my Kaggle profile: Kaggle Profile - heyviv
- Model Improvements: Explore other architectures like EfficientNet-V2 or Transformer-based models.
- Data Augmentation: Experiment with advanced augmentation techniques to further improve performance.
- Deployment: Plan to deploy the classification model using Streamlit or Flask for real-time predictions.
Contributions are welcome! Please fork the repository, submit a pull request, and ensure detailed documentation of any changes.
This project is licensed under the MIT License.