Advanced skin lesion diagnosis with efficientnet-b7 feature extraction and SVM classification

Skin cancer is the most common form of cancer globally. Timely detection is crucial, since failure to identify it in the first stage may result in ‎grave consequences. Notwithstanding its apparent visibility, significant intra-class heterogeneity and inter-class homogeneity complicate its ‎identification. Current AI methodologies for detecting skin cancer are hindered by their reliance on convolutional neural networks, resulting ‎in a lack of interpretability and sluggish processing speeds. To address the issue, the study proposes a comprehensive pipeline that integrates deep learning and machine learning methodologies to enhance detection accuracy in identifying skin lesions. The dataset under con-‎consideration is the International Skin Imaging Collaboration (ISIC) 2020. Initially, we pre-process the photos to guarantee precise training and ‎categorization. The EfficientNet-B7 deep learning model is employed for feature extraction and fed into a support vector machine ‎‎(SVM) classifier. The assessment of parameters such as accuracy, precision, recall, and F1 score yielded an accuracy of 97.52% and an F1 ‎score of 98.61%. The proposed model demonstrates superior results relative to other current models‎.