Skin Lesion 7 Class Classification using MobileNetV2 CNN

The occurrence of skin diseases is widespread across all age groups. These illnesses are more prevalent than other types of health issues. Although traditional diagnostic methods, such as visual examination, biopsy, and microscopic analysis, are available, their accuracy is heavily dependent on skilled professionals or individuals. This paper presents a method for classifying skin lesions utilizing lightweight deep Convolutional Neural Networks (CNNs), with a focus on the MobileNetV2 architecture. Our study employed data up-sampling and data augmentation techniques to enhance the efficiency of the classifier. By utilizing transfer learning with pre-trained models like MobileNetV2, we developed a system that analyzes the HAM10000 dataset, a collection of 10015 dermoscopic images, to achieve an accuracy of 90.30% (categorical), 96.91% (top 2) accuracy, for classifying 7 skin diseases on an Android application. This tool has the potential to assist doctors in detecting skin cancer at an early stage.