Image processing-based skin disease diagnosis: A benchmark study and performance analysis of classifiers

Skin conditions greatly affect an individual's emotional well-being. These illnesses can be devastating to a person's life, as they often require extensive diagnosis and treatment. Skin conditions can change the look and feel of the skin. These illnesses are persistent, contagious, and can sometimes result in skin cancer. Therefore, it is crucial to detect skin conditions early to stop their progression. We propose a technique to detect skin disorders, as delayed diagnosis and treatment of skin infections can cause significant financial and physical strain on the patient. This method involves capturing an image of the affected area, processing it, extracting important features, and using binary classification techniques to make a final determination. Here Convolutional Neural Network (CNN) serves for categorization as well as evaluate its accuracy by implementing four different methodologies for a real-time skin disorder diagnosis system. SVM, CNN, Random Forest (RF), Extra tree (ET) and KNN were compared. It also displays the likelihood of the illness occurring. The Extra Trees classifier achieved the highest accuracy of 96.49% on the given dataset.