Published Paper Details:

Melanoma is a very dangerous kind of skin cancer for which early diagnosis is crucial for successful treatment.Inmost cases,deep learning models such as convolutional neural networks (CNNs) can be employed to classify skin lesions accurately. But becausethey are black-box systems, they cannot be implemented in hospitals since they are not transparent.This paper presents asystemthatintegrates an advanced CNN algorithm named EfficientNet with Explainable AI (XAI) methods to improve accuracy, interpretabilityand transparency in melanoma detection. The model is trained on the ISIC 2016 part 3 dataset, which consist of heterogeneousdermoscopic images of benign and malignant skin lesions. XAI methods offer text explanations of predictions,by indicatingthemostimportant features like asymmetry or border irregularities with audio explanations.The SHAP and LIME are utilized todemarcatetheregions impacted and the contribution of every attributes towards the malignant character. This enhances the process of explanationand informs decision-making. A web application accessible via a user-friendly interface is created using Gradio that allowspatientsand clinicians to upload lesion images for real-time analysis. The web application produces reports, such as predictions,confidencescores and rationales, which are downloadable for use in clinical settings.Comparison is made with a traditional CNNandEfficientNetB3 in accuracy, efficiency, and generalization. Experimental results indicate that EfficientNetB3 has anaccuracyof92.7%, surpassing CNN (84.5%) while retaining computational efficiency. This system solves critical challenges inmelanomadiagnosis by enhancing accuracy and induces trust through interpretability. The principal aim of this project is tominimizeunnecessary biopsies while detecting melanoma and improve early detection leading to improved patient outcomes.