Segmentation based early Melanoma Detection Using Random Forest Algorithm

Abstract

Abstract This article proposes a method to detect melanoma at the early stage before it becomes something severe. The lesion of melanoma has five stages, stage zero to stage four, of which stage one and stage two are early stages. Melanoma is mainly caused by UV radiation, unhealthy lifestyle, hereditary etc.; In addition, age and gender (sex) are also part of the risk factors causing melanoma. So people ought to be aware of what skin disease they have and what precautions and measures they must be taken at its early stage so as to treat it. Because fatal and dangerous cancers must be managed either through prevention or immediate reaction as it occurs. The purpose of this work is to detect melanoma by applying segmentation method at its early stage using random forest classification algorithms. Therefore, segmentation process is implemented as a means of its detection by taking 600 image datasets which are randomly selected from SIIM- ISIC-2020 training dataset. And attained an accuracy of 95%, a precision of 97.3%, a specificity of 95.4%, a selectivity of 95.3% and an f1 score of 96.3. Hopefully, this result will help physicians treat diseases at an early stage and thus prevent further damage.