A Detailed Study on the Choice of Hyperparameters for Transfer Learning in Covid-19 Image Datasets using Bayesian Optimization

For many years, the area of health care has evolved, mainly using medical images to detect and evaluate diseases. Nowadays, the world is going through a pandemic due to COVID- 19, causing a severe effect on the health system and the global economy. Researchers, both in health and in different areas, are focused on improving and providing various alternatives for rapid and more effective detection of this disease. The main objective of this study is to automatically explore as many configurations as possible to recommend a smaller starting hyperparameter space. Because the manual selection of these hyperparameters can lose configurations that generate more efficient models, for this, we present the MKCovid-19 workflow, which uses chest x-ray images of patients with COVID-19. We use knowledge transfer based on convolutional neural networks and Bayes optimization. A detailed study was conducted with different amounts of training data. This automatic selection of hyperparameters allowed us to find a robust model with an accuracy of 98% in test data.

@article{Miranda2021,
  title     = {A Detailed Study on the Choice of Hyperparameters for Transfer Learning in Covid-19 Image Datasets using Bayesian Optimization},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {12},
  number    = {4},
  year      = {2021},
  publisher = {The Science and Information Organization},
  author    = {Miguel Miranda and Kid Valeriano and Jos´e Sulla-Torres},
  doi       = {10.14569/IJACSA.2021.0120441},
  url       = {https://doi.org/10.14569/IJACSA.2021.0120441}
}