A Computational Intelligence Approach for Skin Disease Identification Using Machine/Deep Learning Algorithms

Abstract

Skin diseases are debilitating to a person's health and social life. Among skin diseases, the diagnosis of the most common epidermal diseases would be based on the morphological manifestations of the disease. The morphological manifestations play a key role as an indicator for disease diagnosis. Proper image capture of these symptoms is the first step to diagnosis via the remote process. These images need to undergo pre-processing, and then, the necessary features of the images need to be extracted. This data is then subsequently fed into a model implementation of a machine learning algorithm, which has to be trained to detect and label the skin disease. For this purpose, various algorithms have been implemented over the past decade in the detection of diseases that affect the human epidermis. These algorithms have been noted to achieve varying degrees of success with different epidermal diseases. The most commonly used techniques are Random Forest, Decision Trees, and Naïve Bayes, whereas in recent years, deep learning plays a major role to develop accurate models for skin disease detection. In this chapter, we compare the efficacy of different machine learning algorithm techniques and deep learning techniques that are used for the process of skin disease detection. The algorithms include Random Forest, KNN, Logistic Regression, Naïve Bayes, SVM, and CNN methods along with the performance measures used by the classifiers. To establish the highest accuracy rate possible for diagnosis, a comparison of the results obtained from each algorithm used is necessary. For this comparison, a common dataset with a variety of images is needed. Hence, we propose a novel dataset generated from various sources. The dataset needs to be properly delineated into training sets and testing sets. Here, we also present models that are built to accurately identify five of the most common epidermal diseases with a notably high accuracy rate.