Abstract:
Melanoma skin cancer is increasing rapidly around the world and even in regions such as sub-Sahara where cases have previously been low due to increased ultra-violent radiation. Like many other cancers early detection is vital for survival but unlike many others it occurs on the surface of the skin and not body interior. This makes the use of Computer Aided Diagnosis possible and the most common involves applying ABCDE rule of melanoma to image processing principles. Due to these characteristics high number of features can be derived and it is usually a challenge to find which are the most important for machine learning classification. In this study we apply Random forest algorithm for selecting and ranking of the most suitable features from calculated features for classification based on the ABCDE rule after image segmentation. Statistical evaluation is performed and results show that with few features and random forest classifier, we can achieve comparatively high accuracy.
Keywords:— Melanoma, Skin Segmentationt, Random Forest, Feature Ranking, Feature Selection
