An Ensemble Voting Method of Pre-Trained Deep Learning Models for Skin Disease Identification

Millions of confirmed cancer cases have been reported worldwide as a result of the development of skin disease. One of the most essential stages in preventing disease development is early diagnosis and treatment. Nevertheless, due to similarities in appearance, location, color, and size, diagnosing skin lesions is a challenging feat which requires high standard human resources in the medical system. To address this problem, a machine-based skin disease diagnosis is introduced as a first step to aid in patient classification. Recently, deep learning in medical imaging is becoming a cutting-edge research trend in a variety of applications. In this research, an ensemble network from the pre-trained models ResNet50, MobileNetV3, and EfficientNet is proposed to classify skin diseases. Thanks to the major voting step, the advantages of distinct models are combined to improve the diagnosis of the classification process. The observations and results are based on the experiments performed with the HAM10000 dataset, which includes 7 different forms of skin disease. In comparison to the initial pre-trained models, the proposed model has a 98.3 % average accuracy and other assessment metrics indicate improved results.