Implementation and Evaluation of a Simple Convolutional Neural Network (CNN) for Binary Skin Lesion Classification on the PH2 Dataset

Abstract

This study is essential for automated and accurate classification of skin lesions, main focus of this is to differentiate the benign and malignant tumours cases. Convolutional neuralnet- work(CNN) is build for the study of PH2 dermoscopic imagedataset, this dataset based on almost 430 images, which mainly categories into one of the two main groups: Benign (nevus) and Malignant (melanoma). From the 430 images, 80% data is used for the training and 20% data is used for validation. The model was trained for 10 epochs. After getting results, these results demonstrated high Training Accuracy whose value is 93.91% and also the strong Vali- dation Accuracy whose value is of 77.65%. This all done on unseen test set. There is a huge gap between the confirmed metrics which shows the presence of the overfitting in the data, the one of the common limitation of overfitting is the small size dataset. But , 77.65% validation accuracy help to make a functional and machine learning framework for checking the detection of skin cancer at early stages.