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

Skin cancer, especially melanoma, is one of the most deadly types of cancer. This makes it very important to find it early so that survival rates can go up. Conventional techniques for skin cancer classification often depend on the manual extraction of features, an essential aspect of traditional classification methods, which may encounter limitations due to extraneous patterns and background noise. Convolutional neural networks and other deep learning models have showed a lot of potential in medical imaging, but they often have drawbacks such not being able to be understood effectively and not having enough data to train on. We have introduced the Attention Enhanced Convolution Net (AECNet), an attention-driven convolutional neural network designed to improve feature learning by focusing on the most relevant regions in skin cancer images. This particular concentration enhances the accuracy of classification. To solve the problem of not having enough data, we used an Auxiliary Classifier Generative Adversarial Network (ACGAN) to make fake images, which improved the models’ ability to generalize. Grad CAM was used as an Explainable AI (XAI) technique to explain how the model predicts. This clarifies and builds clinical trust. Testing shows that generated images boost accuracy. The test accuracy of the model is 95.63%. The proposed method enhances diagnostic accuracy and offers a clearer and more dependable answer for practical use.