OESV-KRF: Optimal ensemble support vector kernel random forest based early detection and classification of skin diseases

Abstract

The most widespread disease that often puts the patient’s life in danger is skin disease. Avoiding this situation, it necessitates the development of efficient and reliable skin cancer prediction techniques. Besides, regular screening of suspicious lesions in the skin helps the dermatologist to detect skin cancer in the early phase. Although there exist numerous Machine Learning methods to detect skin cancer in its early stage, the accurate and timely prediction and classification of skin lesions continued to be demanding. To fulfill this aim, in this paper, we proposed an ensemble support vector kernel random forest-based hybrid equilibrium Aquila optimization (ESVMKRF-HEAO) approach. The proposed prediction model is examined using the HAM10000 dataset which contains numerous varieties of skin lesion images in it. Initially, the intrusions and noises in the dataset are removed, and image qualities are enhanced using preprocessing pipelines. Then, the thresholding-based segmentation technique is utilized to segment the cancerous lesion regions from the healthy background. Finally, the proposed classifier accurately predicts and classifies the segmented images based on their feature characteristics into five distinct categories namely melanocytic nevus, basal cell carcinoma, melanoma, actinic keratosis and dermatofibroma. The simulation of the proposed model is performed using MATLAB 2019a software. The evaluation of the proposed ESVMKRF-HEAO approach performance is done concerning metrics such as sensitivity, f-1 score, accuracy, precision and specificity. The proposed ESVMKRF-HEAO approach attained a greater performance rate with respect to all metrics especially, the prediction accuracy of about 97.4% in the experimental results.