Abstract
Skin cancer stands out as the predominant malignancy, necessitating prompt detection and intervention due to its potentially fatal nature. Distinguishing between cancerous and benign skin lesions poses a formidable challenge to visual assessment, underscoring the intricacy of accurate cancer detection. The inherent similarity in the appearances of various lesions further compounds the precision required for effective skin cancer identification. The rapidly evolving technological landscape, notably in the fields of machine learning and deep learning, has seen greater interest in resolving the categorization issues inherent in skin lesions. In our proposed research work, we deploy a deep learning model incorporating a pre-trained DenseNet architecture. This strategic utilization of advanced computational methods aims to enhance the discriminatory capabilities in skin cancer identification. For our research work, we used Melanoma cancer dataset, which contain 10540 dermoscopic labeled images. The study involves extensive preprocessing, including dataset inspection, label encoding, and distribution analysis. The research focuses on training DenseNet-201 architecture, evaluating its performance, and interpreting the results through various metrics.
Keywords: Average Pooling, Classification, DenseNet, Precision, Recall, Skin Melanoma.