Abstract

Skin disease is one of the common illnesses that affect human beings. It affects all cultures, occurs at all ages, and affects between 30 percent and 70 percent of persons. Ineffective classification of skin disease can be caused by bias in the dataset; also, the image background could affect the skin pigment and also affect in carrying out an identification of skin disease. Current progress in pattern recognition has led to success in the development of computerized skin image analysis. In particular, skin disease classification models have gained the feat higher than qualified dermatologists. However, no attempt has been made to assess the steadiness in performance of pattern recognition models across populations with varying skin pigment. In this research, an approach to estimate pigment in benchmark skin disease datasets, and examine if model performance is dependent on this measure. We will also propose the use of two skin disease datasets, one for the Light skin dataset, and it is a collection of dermatoscopic images, and another for the Dark skin dataset which will be obtained locally within Osun State. Feature selection algorithms proposed to be used are Information Gain and Chi-Square, after which we proposed to apply Decision Tree and Random Forest for classification, and then the performance will be evaluated. Our outcome will serve as an evaluation model that will help in improving the accuracy of skin disease detection.