Comparative Analysis of Machine Learning and Deep Learning Based Covid-19 Classification Methods

Al-Areqi F., Konyar M. Z.

8th International Marmara Sciences Congress (Spring 2022), Kocaeli, Türkiye, 13 - 14 Mayıs 2022, ss.1-2

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Kocaeli
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.1-2
  • Kocaeli Üniversitesi Adresli: Evet

Özet

Artificial intelligence has spread to a wide range of applications in recent times. Artificial intelligence is becoming more and more popular as it provides useful solutions for different fields of science since they automate the decision process by extracting the features known as meaningful characteristics in the data. In artificial intelligence applications, machine learning methods in which the features are obtained with the help of various formulas and deep learning methods in which the features are extracted automatically with approaches such as neural networks are used. Because of machine learning and deep learning methods accelerate the decision processes, medicine has been one of the most important usage areas of artificial intelligence. As in many diseases in the field of medicine machine learning and deep learning algorithms give useful results also for coronavirus (Covid-19) disease where rapid diagnosis is crucial. In this study, first of all, public data sets used for the classification of Covid-19 with the help of learning algorithms were examined in detail. Then, current machine learning and deep learning algorithms using these datasets were analyzed. Especially, a comprehensive evaluation has been made on feature extraction methods and classifiers used for machine learning, and current network architectures, transfer learning models, and dataset preprocessing methods used for deep learning. The obtained results were evaluated for disease detection and classification and their usage in the process evaluation of the disease. As a result of this study, we reached two important findings. First, contrary to the classical perception, deep learning methods for Covid-19 do not always achieve better results than machine learning methods. Second, although the accuracy of methods using images from a single device is very good, the accuracy values ​​achieved by methods using datasets collected from different sources or tested on more than one dataset are more realistic.