Automated multi-class skin cancer classification using white shark optimizer with ensemble learning classifier on dermoscopy images

The most prevalent cancer around the world is Skin cancer (SC). Clinical assessment of skin lesions is essential to evaluate the features of the disease; but it is limited by the variety of interpretations and long timelines. A precise and quick diagnosis may assist in increasing the survival rate of the patient. It necessitates the improvement of a computer-aided diagnosis (CAD) system. As accurate and early diagnoses of SC are vital to increasing the survival rate of the patient, deep-learning (DL) and machine learning (ML) methods were introduced to support dermatologists and resolve the abovementioned problems. This manuscript introduces an Automated Multi-class Skin Cancer Classification using White Shark Optimizer with Ensemble Learning (AMCSCC-WHOEL) Classifier model on Dermoscopy Images. The presented AMCSCC-WHOEL technique exploits majority voting ensemble DL classifier models with a hyperparameter tuning strategy for the classification of SC. To accomplish this, the AMCSCC-WHOEL technique comprises image preprocessing in two stages: contrast enhancement and high-frequency filtering (HFF) based noise removal. Besides, the ShuffleNet-v2 model is used for the generation of feature vectors. For SC classification, a majority voting ensemble classifier comprising three DL models is used, namely sparse autoencoder (SAE), deep convolutional autoencoder (DCAE), and attention-based bidirectional gated recurrent unit. WHO method is applied for optimal hyperparameter tuning process to enhance the detection performance of the ensemble models. The experimental analysis of the AMCSCC-WHOEL algorithm is tested on a benchmark dataset. Extensive result analysis pointed out the improved detection results of the AMCSCC-WHOEL method over other recent approaches.