Performance of Optimization Techniques in the Deep Learning Approach for Classification of Covid-19 Pneumonia

This study concerns the COVID-19 pneumonia classification based on X-ray images of the lungs using a deep learning-based approach. One of the crucial parts of the deep learning approach is the optimization process. Thus, determining the optimization technique precisely is crucial to producing optimal performance from the deep learning approach. The problem is that there is no single optimization technique that produces the best performance in all cases. In this study, various optimization techniques are tested to evaluate their effectiveness in the classification of covid-19 pneumonia. The approach in the study uses the CNN method. Three categories are used to carry out the classification, namely covid19 pneumonia (inflamed lungs due to covid-19), pneumonia (inflamed lungs due to factors other than Covid-19), and normal (healthy lungs). The optimization techniques tested included RMSprop, Adadelta, Adam, Adamax, and Nadam. In the experimental process, we test the method on a lung image dataset consisting of 3,900 training images, 450 validation images, and 225 test images. The result of this study concludes that the performance of the Adamax technique outperforms other optimization techniques. The evaluation of the Adamax technique result in accuracy, precision, recall, and f1 score of 95.11%, 0.9596, 0.9511, and 0.9553, respectively.