Skin disease classification system based on metaheuristic algorithms

Abstract

Some doctors face difficulty in diagnosing certain types of skin diseases due to the high similarity among them. Six skin diseases are an example of such types:(seborrheic dermatitis,psoriasis, lichen planus, chronic dermatitis,pityriasisrosea,pityriasisrubra) and collectively called Erythemato-Squamous Diseases. Automated systems for classifying skin diseases have emerged as a result of the phenomenal advancement of computer technology in all areas of life. Automated, effective, and accurate classification of skin diseases is very important for biomedical analysis. Sine Cosine Algorithm (SCA) is one of the modernmetaheuristic algorithms proposed to solve many optimization problems. This research paper offers a new Feature Selection (FS) approach that proposes converting the original SCA to a Binary version (BSCA) for applying in the classification domain to determine the best feature subset based on the wrapper model. Thereafter, Enhancing BSCA by the mutation operator to produce a hybrid approach (ESCA). The mutation is entered as an internal mechanism to preserve diversity and strengthening the SCA's exploration capabilities. The result obtained from the proposed approach ESCA was compared with the BSCA and other FS approaches such as Antlion Optimization Algorithm (ALO), and Particle Swarm Optimization (PSO). The dataset for testing was obtained from the UCI Repository site, it consists of 366 samples with 34 features. The experimental results demonstrate the effectiveness of the suggested approach in extracting optimum features from among the overall features of the dataset. ESCA gave excellent diagnostic accuracy (0.981410) with the ratio of selected features (0.564706).