Enhanced Skin Cancer Classification: Proposed Efficient Combination Model in CNN Architectures

Skin cancer is the abnormal and uncontrolled growth of cells in the body, often linked to prolonged exposure to harmful ultraviolet rays from the sun. Studies indicate that distinguishing between benign and malignant skin lesions is critical for ensuring proper patient care. Evidence also suggests that, compared to traditional approaches, convolutional neural networks (CNNs) are significantly more effective in skin lesion classification tasks. This research presents a suitable hybrid deep convolutional neural network (CNN) architecture designed to enhance the identification of skin cancer. The proposed approach leverages the strengths of two primary heuristics, combining the lightweight yet powerful feature extraction capabilities of EfficientNetB0 with ResNet50, as well as EfficientNetB0 with DenseNet121. To further optimize the performance of the models and address the inherent imbalance often found in medical datasets, comprehensive data augmentation techniques were employed. These techniques not only improve the diversity of the training data but also enhance the model's ability to generalize effectively to unseen data, thereby ensuring more reliable and accurate predictions. This study uses the ISIC datasets to evaluate the proposed approach. The results show that the combination of ResNet50 and EfficientNetB0 achieves an accuracy of approximately 0.95%, while DenseNet121 and EfficientNetB0 attain a significantly higher accuracy of 97.8%. These findings indicate that the performance of the proposed method surpasses that of existing skin cancer classification techniques, highlighting its effectiveness and potential for improved diagnostic accuracy.