Ensemble Learning for Detection of Types of Melanoma

Abstract

Melanoma is a potentially fatal type of skin cancer in these melanocytes develop uncontrollably. Malignant melanoma is another name for melanoma. Melanoma rates in Australia and New Zealand are the highest in the world. Melanoma is anticipated to strike one in every 15 white New Zealanders at some point in their lives. Invasive melanoma was the third most prevalent malignancy in both men and women in 2012. Melanoma can strike adults of any age, but it is extremely uncommon in youngsters. Melanoma is hypothesised to start as an uncontrolled proliferation of genetically transformed melanocytic stem cells. Early diagnosis of melanoma in Dermoscopy pictures boosts the survival percentage substantially. Melanoma detection, on the other hand, is extremely difficult. As a result, automatic identification of skin cancer is extremely beneficial to pathologists' accuracy. This paper offers an ensemble deep learning strategy for accurately classifying the kind of melanoma at an early stage. The proposed model distinguishes between lentigo maligna, superficial spreading and nodular melanoma, allowing for early detection of the virus and prompt isolation and treatment to prevent the disease from spreading further. The deep layer architectures of the convolutional neural network (CNN) and the shallow structure of the pixel-based multilayer perceptron (MLP) are neural network algorithms that represent deep learning (DL) technique and the classical non-parametric machine learning method. Two methods that have diverse behaviours, were combined in a simple and successful means for the classification of very fine melanoma type detection utilising a rule-based decision fusion methodology. On dataset retrieved from https://dermnetnz.org/, the efficiency of ensemble MLP-CNN classifier was examined. In compared to state-of-the-art approaches, experimental outcomes reveal that the proposed technique is worthier in terms of diagnostic accuracy