Ensemble learning methods for deep learning: Application to skin lesions classification
Amina Aboulmira, El Mehdi Raouhi, Hamid Hrimech, Mohamed Lachgar
Abstract
Despite all the contributions of deep learning (DL) which have contributed to improving performance in the field of dermatology, the classification of skin diseases remains challenging. Indeed, deep learning has made a great advancement in the field of medicine and has had the potential to improve practice in this field, from diagnosis to personalized treatment. However, it is not always relevant for certain issues, the performance of a single DL model is not always satisfactory. Ensemble approaches have proven their effectiveness in solving many Deep Learning challenges such as the classification task, as well, the accuracy of the final prediction process can be improved by training many models and combining them. In this paper, the ensemble learning approach is explored by combining several architectures in order to detect and classify benign and malignant skin cancer diseases. The results obtained demonstrate the usefulness of this approach in the classification of skin cancer images, ensemble model achieved an accuracy rate 0.96 of and AUC score of 0.96.