AECNet: Advancing Skin Cancer Classification using ACGAN-Based Synthetic Data with Grad-CAM Explainability

One of the most deadly types of cancer is skin cancer, especially melanoma; early discovery is very important to increase survival rates. Handmade feature extraction, a crucial component of conventional approaches for skin cancer classification, may face constraints due to irrelevant patterns and background noise. Although deep learning models such as convolutional neural networks (CNNs) have shown potential in medical imaging, they often lack interpretability and suffer from data shortages. Attention Enhanced Convolution Net (AECNet) is a hybrid attention-based CNN model proposed in this work to improve feature learning by concentrating on pertinent areas of skin cancer images, hence raising classification accuracy. We used an Auxiliary Classifier Generative Adversarial Network (ACGAN) to create synthetic images, thereby improving the generalizing capacity of the model in order to solve data shortage. We also incorporated Grad-CAM Explainable AI (XAI) methods to offer visual explanations of the model's decision-making process, hence enhancing its clinical relevance and interpretability. Using synthetic images, findings show notable increases in classification accuracy, precision, recall, and F1-score; the model achieves a test accuracy of 95.63%. Apart from improving the accuracy of skin cancer diagnosis, the suggested method offers a more open and understandable structure for clinical uses.