Performance Analysis of Various Feature Extraction Methods for Classification of Pox Virus Images

Abstract

After the COVID-19 pandemic, people began to fear that the monkeypox virus will be the next epidemic. The World Health Organization (WHO) has reported that monkeypox outbreaks have taken place in different regions of Central and West Africa throughout the years, with the latest outbreak being reported in Nigeria in May 2021. Fever, swollen lymph nodes, dry cough, and red rashes all manifest as signs of the monkeypox virus. Most of the symptoms of Measles and chickenpox are comparable. These disorders are given only methodical treatment by the doctor. In Image processing, feature extraction techniques are used to transform raw pixel values of an image into a set of features that can be used for further analysis, such as object recognition, image classification, and image retrieval. Several feature extraction techniques are employed to determine the disease from the images in order to determine which technique works best for this collection of monkeypox skin images (MSID). Wavelets fused with gray-level co-occurrence matrix (GLCM), Haralick features, and local binary pattern are the various feature extraction techniques (LBP) applied in this work. For the classification of various pox virus diseases such as measles, chicken pox, and monkeypox, various classification algorithms such as Random Forest Classification (RF), Naive Bayes (NB), K-Nearest Neighbor algorithms (KNN), Support Vector Machine (SVM), Ada Boosting (AB), and Gradient Boosting (GB) are used in this work. In this paper, four evaluation metrics are used to determine the best feature extraction method for the monkeypox, chickenpox, and measles datasets. Wavelets fused with GLCM produce the highest accuracy (84.41% for gradient boosting and 83.87% for random forest) when extracting features from Monkeypox Skin Image Datasets (MSID).