Channeling Fairness: Class Imbalance-Aware Skin Disease Recognition via Fair Channel Enhancement Module

This project develops a fair skin disease classification system using ResNet50 and SE-EfficientNet-B3 with a Fair Channel Enhancement Module (FCEM) to dynamically reweight features for minority classes, addressing dataset imbalance on the Kaggle skin disease dataset. A Flask-based web interface with HTML, CSS, and JS allows user registration, login, and image submission for real-time predictions. Performance is evaluated per class using accuracy, precision, recall, and F1-score, comparing architectures to identify the most effective and fair backbone. The design emphasizes modularity, robustness, and reproducibility, enabling easy dataset expansion and future integration. Ultimately, it demonstrates fairness-aware AI for medical image classification, mitigating prediction bias across all disease categories.