Confirming Covid-19 Pneumonia Based on Chest X-ray Classification

Abstract

Rapid and affordable diagnostic tools are essential for detecting COVID-19 induced pneumonia, particularly in healthcare systems with limited resources. This study proposes a machine learning and deep learning framework to confirm COVID-19 pneumonia using chest X-ray images. A publicly available Kaggle dataset containing 6,432 X-ray images categorized into COVID-19, pneumonia, and normal cases was used. The dataset was divided into 80% training and 20% testing subsets, with 5-fold cross validation applied to ensure evaluation robustness. Six models were developed three conventional classifiers Support Vector Machine, Random Forest, Naïve Bayes and three convolutional neural networks VGG16, DenseNet-121, and InceptionV3. Image preprocessing included normalization and controlled augmentation to enhance model generalization while preserving clinical realism. Model optimization involved grid search for hyperparameter tuning and adaptive learning rate callbacks to prevent overfitting. Among the tested architectures, DenseNet-121 achieved the highest accuracy of 94.17%, outperforming comparable studies on the same dataset (typically reporting 90-92% accuracy). This improvement demonstrates the effectiveness of the optimized training pipeline and balanced augmentation strategy. The proposed approach is computationally efficient and relies entirely on open source datasets and models, aligning with the goal of developing affordable and scalable Al assisted diagnostic systems for medical imaging. Future work will explore lightweight CNN architectures and ensemble learning to further enhance generalization and interpretability.