Skin Cancer Classification and Early Detection on Cell Images using Multiple Convolution Neural Network Architectures

Skin cancer is an abnormal growth of epidermal cells that can spread over time. It can develop gradually and become death threatening if left untreated. Early detection and prevention are essential as it may prevent the cells from spreading. In this study, skin cell image data is used to classify Benign and Malignant cells using CNN models. The image dataset goes through several data preprocessing techniques before being applied to achieve the most efficient outcome from four pre-trained CNN models. The pre-trained models used in the study are InceptionV3, MobileNetV2, VGG19, and EfficientNetB7. With the advancing epoch, an increase in accuracy is observed from the models. The highest accuracy of InceptionV3, MobilenetV2 and VGG19 is 89.86%, 83.06% and 77.5%, respectively, whereas EfficientNetB7 comes with a significantly lower accuracy score of 49.03%. The domination of InceptionV3 over all other models is observed in Precision, Recall and F1-Score while classifying skin cancer cells.