ARTIFICIAL INTELLIGENCE APPROACH FOR AUTOMATIC MULTICLASS SKIN DISEASES IDENTIFICATION

Skin disease is one of the most frequent health problems. Most of the skin diseases are not malignant, hence it is mostly ignored. In fact, the unappropriated treatment will potentially reduce the quality of life. On the other hand, delivering skin disease diagnostics is challenging. Diagnosis error is evitable due to some of the symptoms' similarities. In recent years, Artificial Intelligence (AI) approach has been promoted to support medical practice to deliver accurate skin diseases diagnosis. Most of the works are still focusing on detecting the deadliest skin disease, Melanoma. There are limited works on exploring AI for more general multi class skin diseases identification. This paper presents the simulation works and analysis on Machine Learning approach for detecting general multi class skin diseases. In addition, to provide comprehensive insight about the AI utilization in dermatology, the research works mapping of AI approach on dermatology is also presented. On the simulation works, two machine learning frameworks, traditional machine learning and transfer learning approach are employed, analysed, and discussed. Six (6) different skin diseases, dermatofibromas, chickenpox, eczema, keratosis, psoriasis, and scabies are used as the study cases. Based on the simulation results, the proposed methods achieve 70%-80% accuracy. The traditional machine learning approach shows competitive performance to the transfer learning approach. Based on the accuracy metric and convergence behaviour, both of these approaches are promising to be further developed and modified with bigger datasets.