<![CDATA[Melanoma skin cancer is one of the most dangerous skin cancers where the ferocity and speed of metastasis has caused a high mortality rate among afflicted when the cancer is not treated. Early detection of the cancer and prevention by removing the affected skin have been shown to decrease the mortality rate on afflicted patient. Thus, development of a method to help automatically diagnose the cancer and classify between cancer and normal mole or birthmark is needed. Previous methods still show limitations in classifying melanoma skin cancer. This study proposes a classification system using convolutional neural network trained on the original ISIC 2020 dataset and hair removed dataset which is then combined using ensemble. The dataset used is first preprocessed using the hair removal algorithm convolutional neural network using EfficientNet B0 – B7 and ResNet-50-v2 will be trained using ISIC 2020 dataset and ISIC 2020 dataset processed with hair removal algorithm.The model is evaluated using test data from ISIC 2020 dataset on area under the receiver operating characteristic curve (ROC AUC). The model trained will then be combined using ensemble where the result of the model will be averaged to give a combined prediction. The result of the test shows that the model trained is capable to classify melanoma and non-melanoma images. It is also shown that by removing hair from the skin image reduces the accuracy of th e model. Using Ensembling on the different models trained into one meta-model also increases the accuracy of the prediction giving a high final accuracy of 93.108%.]]>
