Machine Learning Model for Multiclass Lesion Diagnoses

Abstract

Cancer detection is one of the most important research fields in the area of intelligent computing. Skin lesion diagnosis is a challenging topic, and several models have experimented on different datasets. Researchers proposed classification models that classify the lesion type if it is malignant or benign. The aim of this research is to propose a multiclass machine learning model that detect the lesion diagnosis rather than its type. The used dataset was retrieved from the International Skin Imaging Collaboration datasets archive since it is a benchmark that has thousands of dermoscopic images of different diagnoses. Melanoma, Nevus, and Seborrheic keratosis were the used lesion diagnosis. The proposed model consists of sequential phases, that start with the filtering and end with the classification. Kernel Support Vector Machine and Random Forest were the classifiers of the proposed model and their performance was measured by the KFold cross-validation accuracy.