Optimizing Pneumonia Detection from Chest X-Rays: A Performance Comparison of Lightweight Cnns and Ensemble Methods

Pneumonia remains a critical global health concern, necessitating rapid and accurate diagnostic strategies to improve patient outcomes. This study evaluates lightweight deep learning models and ensemble learning techniques for pneumonia detection from chest X-ray images. Five fine-tuned convolutional neural networks (CNNs)-MobileNet, EfficientNetB0, DenseNet121, NASNetMobile, and ResNet-were optimized using a curated dataset. A soft voting ensemble was also employed to combine model predictions. However, the best-performing individual models, MobileNet and EfficientNetB0, outperformed the ensemble across key metrics, achieving accuracies of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 5. 1 \%}$</tex> and an AUC of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 9. 7 \%}$</tex>, respectively. These results highlight the robustness of lightweight CNNs as standalone diagnostic tools, with ensemble learning providing limited added value. This study offers practical insights into model selection for early and accurate pneumonia diagnosis in medical imaging.