Sheetal Nana Patil, Prashant G Patil & itendra Prakash Patil – Assistant Professors
R C Patel Institute of Technology, Shirpur (M.S) India.
Abstract
the fourth greatest source of non-fatal diseases affecting 30-70% of the population worldwide is skin disease, and it is
common across geography and ages. Skin disease is the predominant form of skin cancer in the United States. Skin disease is one of
the most common health problems. In order to provide statistical representations on suspended areas, computer-operated diagnostic
systems require sophisticated image processing algorithms. In this article we analyzed the latest scientific progress used for the
diagnostic methods for computer-aided skin lesions. The phases include pre-processing of skin lesions, segmentation, collection of
features and the recognition of particular characteristics. In this paper we implemented Random Forest (RF) classification algorithm
to classify benign and malignant small skin image dataset. In RF classifier Bootstrap aggregation and selection of K and N node is
key stage which decide performance of algorithm. Classifier Performance was improved in terms of accuracy (+3.44%), Area under
ROC (+1.05%), precision (-0.8%) during cross validation score. Results show that RF classifier is able to achieve classification
accuracy 86.90% for malignant cases and similarly 88.67 % for benign cases.