WEL-ODKC: weighted extreme learning optimal diagonal-kernels convolution model for accurate classification of skin lesions

Abstract

ABSTRACT This paper presents a new model, the Weighted Extreme Learning Machine optimized Diagonal-Kernels Convolution (WELM-ODKC), for automatic skin cancer detection that addresses imbalanced data and eliminates inter-operator variability. The model combines the WELM and the Enhanced Remora Optimization Algorithm (EROA) with the Diagonal-Kernels Convolution Neural Network (DKCNN) to enhance the weight function and accurately predict skin lesion class. The model was evaluated on the MNIST HAM10000 and PAD-UFES-20 datasets and outperformed other existing skin cancer classification methods such as SVM BWO, CNN, DGC-NB, and GWO-CNN. The accuracy, recall, precision, and F1-score were used to evaluate the performance of WEL-ODKC, and the results show a high accuracy during training and validation. The proposed model efficiently classifies different types of skin cancer images from the two baseline datasets and provides promising results for automatic skin cancer detection.