ML and DL architecture comparison for COVID-19 classification using Chest X-ray Images

The documentation for the whole approach can be found in the following published book chapter:

• Afzal, E., Saba, T., Ali, H., Karim, A., Javed, K. (2022). ML and DL Architectures Comparisons for the Classification of COVID-19 Using Chest X-Ray Images. In: Saba, T., Rehman, A., Roy, S. (eds) Prognostic Models in Healthcare: AI and Statistical Approaches. Studies in Big Data, vol 109. Springer, Singapore. https://doi.org/10.1007/978-981-19-2057-8_16

The overall Approach

The whole study can be divided into the following sections:

1. Data Collection

The data used in this study can be found on Kaggle through the folowing link: COVID-19-Pneumonia-Normal-Chest-Xray-Dataset

2. Data Preprocessing

The whole dataset was divided into 2 categories, as follows:

1. Normal
2. COVID-19

The dataset of the pneumonia X-rays was omitted for the study. After the split, a number of preprocessing and augmentation techniques were used to introduce variation in the data and also to populate it.

3. Model development

The study was conducted so as to compare both machine and deep learning models. The machine learning models used in the study include:

1. Support Vector Machine
2. Logistic Regression
3. Decision Tree

The code for the machine learning models can be found in the Machine_learning_models folder.
Apart from this, the deep learning models used include convolutional neural networks, both with self defined layers and pre-trained models used for transfer learning. The deep learning models used include:

1. Convolutional Neural Network (Self Defined)
2. VGG-19
3. ResNet-50
4. AlexNet

The Convolutional Neural Networks (CNN's) code is contained in the folder Convolutional_neural_network, while the other three models are located in Transfer_learning_models.

4. Model Evaluation

Multiple evaluation metrices have been used for the evaluation of these models. The metrices we used in the study are:

1. Accuracy
2. Precision
3. Recall
4. F1-measure
5. Area Under Curve (AUC)