Performance of Root-Mean-Square Propagation and Adaptive Gradient Optimization Algorithms on Covid-19 Pneumonia Classification

ABSTRACT

The SARS-CoV-2 coronavirus causes inflammation of the lungs, known as Covid-19 Pneumonia. Doctors or radiologists usually use lung images from X-rays to detect the condition of a person's lungs has Covid-19 Pneumonia or not. This research classifies x-ray images of the lungs using deep learning inti 3 categories, namely Covid-19 Pneumonia, Ordinary Pneumonia, and Normal. This method for classification uses the Convolutional Neural Network (CNN), which applies 22 layers containing 5 Convolutional Layers with dimension values 16, 32, 64, 128, and 256. This research tested the Root-Mean-Square Propagation (RMSprop) and Adaptive Gradient (Adagrad) optimization algorithms used to optimize the CNN performance model for Covid-19 Pneumonia classification. The experiment involved 3.900 lung images for the training process, 450 lung images for validation, and 225 lung images for testing. Based on the investigation, implementing the RMSprop optimizer produces an accuracy of 87.99%, a precision of 0.88, a recall of 0.86, and an f1 score of 0.87. Meanwhile, implementing the Adagrad optimizer produces an accuracy of 75.99%, a precision of 0.79, a recall of 0.72, and an f1 score of 0.75. These results provide essential information that the optimization algorithm of the RMSprop produces better performance than the Adagrad in classifying Covid-19 Pneumonia. © 2022 IEEE.