International Research Journal of Education and Technology

Skin cancer is one of the most common types of cancer, and its early detection can be life-saving. In this project, we built a system that uses deep learning to classify skin lesions as either benign or malignant. We applied three well-known models- InceptionV3, Xception, and EfficientNet to achieve this. While these models are highly accurate, they often operate as "black boxes," making it hard for doctors to understand how they reach their decisions. To solve this, we integrated explainability techniques like Grad-CAM and SmoothGrad, which visually show which areas of the image influenced the model’s decision. This not only improves trust in the system but also makes it easier for healthcare professionals to verify its results. Our experiments showed that Xception performed best in terms of both accuracy and explanation quality. This work brings us closer to making AI-driven tools more trustworthy and transparent in real-world medical applications.

Keywords: Skin Lesion Classification, Deep Learning, InceptionV3, Xception, EfficientNet, Explainable AI, Grad-CAM, SmoothGrad, Medical Image Analysis, Skin Cancer Detection.