Transfer Learning Models Comparison For Detecting And Diagnosing Skin Cancer

Skin cancer is a severe problem that is frequently disregarded. In circumstances of manual examination by a clinician, the human eye is occasionally unable to detect disorders precisely from imaging data. Deep learning techniques are increasingly being used nowadays to solve various problems in our daily lives. Therefore, deep neural network techniques are used to create an automated and computerized mechanism for detecting skin illnesses. To identify and diagnose skin illnesses over a range of criteria several neural network algorithms are evaluated and tested in the suggested model to see how well they perform. The networks are constructed to provide better outcomes using the CNN (Convolution neural network) and the Keras Sequential API architectures. The paper also compares the outcomes of the models using several metrics, such as accuracy, precision, f1 score, and recall. The transfer learning model involves seven models like DenseNet201, InseptionResnetV2, MobileNetV2, InceptionV3, ResNet50, DenseNet169, and VGG16. Among the employed models, the DenseNet169 model achieved the highest score of 87.58% in terms of accuracy; also, in terms of sensitivity and F1 score, DenseNet201 achieved the highest scores of 95.28% and 89.09%, respectively. On the other hand, VGG16 gained a score of 89.67% in terms of specificity, and DenseNet169 achieved the highest score of 90.64% in terms of precision.