Pigmented Skin Lesions Classification Using Data Driven Subsets of Image Features

In this paper we present an architecture for identification of pigmented skin lesions from dermatoscopic images. The architecture used a large number of image features and was evaluated with several classification algorithms on different feature subsets as extracted from feature ranking. The best performing classification algorithm was the support vector machines using polynomial kernel function with classification accuracy equal to 74.69% and the most precisely classified skin lesion type between seven different skin pathologies was nevus with accuracy equal to 94.38%.