Melanoma Classification using XGB Classifier and EfficientNet

Melanoma is the most prevalent occurring cancer. Currently, dermatologists evaluate the patient's moles individually to identify outlier lesions most likely to be melanoma. In recent years deep learning techniques have shown promising results in various medical classification challenges, such as skin lesion classification. Usually, in these challenges, the classification is done by training a deep neural network on the image dataset for prediction. This paper proposes a two-stage architecture where stage 1 predictions are based on XGB Classifier with adversarial validation, and stage 2 is based on EfficientNet(EffNet) trained on high-resolution images in TFRecords format. The main objective is to propose a superior method in terms of classification accuracy and lower computational costs.