Dermatological Cancer Classification

Early and accurate detection of skin cancer is critical for effective treatment and improved patient outcomes. In this study, we develop and evaluate deep learning models, including DermaNet, ResNet50, EfficientNetB0, and a hybrid CNN, for dermatological cancer classification. The HAM10000 dataset was preprocessed using data augmentation and oversampling techniques to address class imbalances. Additionally, a "normal" class was introduced to filter out non-relevant images, enhancing the model's real-world applicability. DermaNet achieved the highest classification accuracy of 97.14%, followed by ResNet50 (95.74%) and EfficientNetB0 (88.50%), while the hybrid CNN effectively distinguished skin cancer from non-skin cancer cases with 91.87% accuracy. The study further explores model interpretability, hyperparameter tuning, and computational efficiency, making it suitable for real-time clinical deployment. These findings highlight the potential of deep learning for automated skin cancer diagnosis and emphasize the need for further validation on diverse datasets.