Publication

Abstract : Pneumonia continues to be a significant global health challenge, emphasizing the need for precise detection methods. This study explored a weighted ensemble model that combines ConvNeXt, Swin Transformer, and EfficientNetV2 for enhanced analysis of chest X-ray scans to facilitate more accurate pneumonia detection. By integrating both convolutional neural networks (CNNs) and vision transformers (ViTs), our approach captured richer features, leading to improved accuracy and generalization. Our model achieved 97.12% test accuracy, surpassing the previous benchmark of 93.8%, highlighting the benefits of ensemble learning in medical imaging. To ensure a thorough evaluation, we assessed the performance using precision, recall, F1 metric, and ROC-AUC assessment. Experimental evaluations were conducted using Google Colab’s T4 and A100 GPU to ensure efficiency and scalability. The results demonstrate that ensemble learning strengthens the classification performance, making AI-driven pneumonia detection more reliable and accessible. This approach is particularly valuable for resourcelimited healthcare settings where expert radiologists may not always be available. By providing a scalable and efficient solution, our work contributes to translating advancements in AI research into practical, real-world clinical solutions, ultimately helping to improve pneumonia diagnosis and patient care.