Skin Cancer Detection using Improved Empirical Wavelet Transform Features with SVM Classification

  • G.Neela Krishna Babu, Dr.V.Joseph Peter
Keywords: Skin Cancer Detection, Random Sampling, Improved Empirical Wavelet Transform (EWT), Grid Search, Support Vector Machine (SVM).

Abstract

This research work proposes an improved Empirical Wavelet Transform (EWT) features with grid search based Support Vector Machine (SVM) classification technique to identify skin cancer in dermatoscopic images. All images from the International Skin Imaging Collaboration 2018 (ISIC 2018) dataset contain an imbalanced set of classes in seven cancer types. Random sampling technique is performed over these imbalanced set of classes to rebalance class labels. The EWT algorithm is then processed in these rebalanced images to decompose the input signal into approximation and several detailed coefficients. Here, the approximation coefficient is further processed into grid search based SVM classification for non-linear separation of classes from dataset to find seven skin cancer types. Thus, the proposed method achieves 95% Accuracy, 91% Specificity, 95% Precision, 95% F1-score and 95% Recall values, which is competitively higher than the existing skin cancer methods.

Published
2021-09-17
How to Cite
Dr.V.Joseph Peter, G. K. B. (2021). Skin Cancer Detection using Improved Empirical Wavelet Transform Features with SVM Classification. Design Engineering, 12267-12278. Retrieved from http://www.thedesignengineering.com/index.php/DE/article/view/4388
Section
Articles