COVID-19 detection using chest X-ray images based on Machine learning and Deep learning models - Further evidence from Data Augmentation

Aug 4, 2026

Upload by Finn

Rapid, accurate COVID-19 detection was critical for pandemic control through timely quarantine and medical care, especially when vaccines were unavailable and detection kits were limited. This study develops a diagnostic tool using accessible resources and advanced deep-learning techniques. Two chest X-ray dataset sources with normal, COVID-19, and viral pneumonia images were used. Feature extraction supported Random Forest and SVM classification, while data augmentation improved performance on small and reduced images for imbalanced data. Deep models (VGG19, CNN, convolutional autoencoder) with hyperparameter tuning and transfer learning yielded higher accuracy, though overfitting could reduce generalizability.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages83

Reading Time3 h 29 min

AccessFree to download

Abstract

Introduction

Background

Motivation for the present research work

Organization of the thesis

Literature Review

Traditional Machine Learning Models

Deep Learning Based Methods

Hybrid Methods

Methodology

Random Forest

Support Vector Machine (SVM)

Convolutional Neural Networks (CNN)

VGG19

Convolutional Auto Encoder

Data

Data Augmentation

Dataset 1

COVID-19 detection using chest X-ray images based on Machine learning and Deep learning models: Further evidence from Data Augmentation Submitted in partial fulfilment of the requirements of the degree of Master of Data Science Shokoufeh Hosseini (Student Number: 116276) Supervisor: Associate Prof. Habib Ullah Co-Supervisor: Associate Prof. Fadi Al Machot Faculty of Science and Technology Norwegian University of Life Science May 2023 Declaration This written submission represents my own ideas in my own words, and where others’ideas or phrases have been used, the original sources have also been appropriately cited. Furthermore, I declare that my submission complies with all academic integrity and honesty principles and does not contain any misrepresentation or fabrication of any idea, fact, or data. The university can take disciplinary action against those who violate the rules. This action can also elicit penalties from sources without adequately cited or whose permissions would have been obtained. Shokoufeh Hosseini Date: May 14, 2023 CERTIFICATE It is certified the work contained in the thesis titled “ COVID-19 detection using chest Xray images based on Machine Learning and Deep Learning models: Further evidence from Data Augmentation”by “ Shokoufeh Hosseini”has been carried out under my supervision and that this work has not been submitted elsewhere for a degree. Dr. Habib Ullah Associate Professor, Dept. of Science and Technology Norwegian University of Life Science Dr. Fadi Al Machot Associate Professor, Dept. of Science and Technology Norwegian University of Life Science Date: May 14, 2023 Place: Norwegian University of Life Science ACKNOWLEDGMENTS I am delighted to express my respect and a deep sense of gratitude to my Master supervisor Habib Ullah & my Co-supervisor Fadi AL Machot Associate Professors, Department of REALTEK for their wisdom, vision, expertise, guidance, enthusiastic involvement, and persistent encouragement during the planning and development of this research work. I also gratefully acknowledge their painstaking efforts in thoroughly reviewing and improving the manuscripts, without which this work could not have been completed. I am incredibly thankful to my husband Sajad for his knowledge, love, and encouragement during this journey. I wish to express my appreciation to my friends and thanks to the research fellows at the department for their help and motivation throughout my research work. I also would like to express my deep and sincere gratitude to my friends and all other persons whose names do not appear here. These people have directly or indirectly helped me in good times and hard times. Finally, I am indebted and grateful to the Almighty for helping me in this endeavor. Shokoufeh Hosseini ABSTRACT Detecting COVID-19 Coronavirus quickly and accurately was essential for preventing and controlling this pandemic through timely quarantine and medical treatment in the absence of vaccines. Because there were an increasing number of cases ofCOVID-19 worldwide and the limited number of detection kits available, it was difficult to identify the presence of this disease. Consequently, we needed to seek other alternatives at this time. Deep learning techniques in computer-aided medical diagnosis have surpassed state-of-the-art performance in computer-aided diagnosis among existing, widely accessible, and low-cost resources. This dissertation proposes an alternative diagnostic tool that utilizes available resources and advanced deep-learning techniques to detect COVID-19 cases. We used two sources of datasets in this study, one of which was a small dataset [1], and another was a large dataset [2] with X-ray images of normal, COVID-19, and viral pneumonia. Using chest X-ray images, we investigated the performance ofCOVID-19 detection using machine learning and deep learning methods. We extracted edge and pixel values using feature extractio

FAQs

What problem does the thesis address?

The thesis targets the challenge of detecting COVID-19 quickly and accurately when detection kits are limited and case numbers rise. It also aims to improve diagnostic performance using X-ray based machine learning approaches.

Which datasets and classes are used for training and evaluation?

Two sources of X-ray image datasets are used, containing normal images, COVID-19 images, and viral pneumonia images. Dataset size differences and class imbalance are addressed using augmentation and image reduction.

How do the proposed models make predictions, and what improves performance?

Traditional models extract edge and pixel-based features and classify with Random Forest and SVM. Deep learning models (VGG19, CNN, convolutional autoencoder) use hyperparameter tuning and transfer learning, and data augmentation (brightness, contrast, flipping) to enhance results.

Aug 4, 2026

Upload by Finn