Feature selection using sequential backward method in melanoma recognition

Due to increased ultraviolet radiation, melanoma skin cancer is on the rise globally even in darker-skinned communities, new cases are being discovered. When detection is in the early stages of cancer like many other forms of cancers, the chances of successful treatment and cure are higher. Chance of survival reduces significantly if detected at a later stage. Diagnosis of melanoma using image processing systems have been developed to assist dermatologist. These employ image processing techniques based on the ABCD rule of melanoma to perform feature extraction after segmentation of the affected region to derive characteristic that will enable machine learning algorithms to classify cancerous from the non-cancerous skin. This paper suggests the use of sequential backward selection technique to determine what are the least number of features that can be used for high accuracy in machine learning classification using k-Nearest Neighbors algorithm.