A Hybrid Ensemble Deep Learning Framework for Pediatric Pneumonia Classification Using Transfer Learning and Convolutional Neural Networks

The current implementation of pneumonia diagnosis remains challenging to achieve better performance and improve results. The aim of this research is to propose an innovative framework for pediatric pneumonia diagnosis that unites three fine-tuned pre-trained CNN models through feature fusion at the EfficientNetB0, RestNet50, and MobileNetV2 to achieve better performance and results. The mixed-model architecture framework provides an ideal solution for time-sensitive clinical applications operating in resource-constrained environments. This research experiment used the Chest X-Ray Images (Pneumonia) dataset, which contains 5863 high-resolution anterior-posterior (AP) chest radiographs sampled from children aged 1 to 5 years old. This study presents four key novelties. Firstly, we systematically evaluated five CNN (Convolutional Neural Networks) combinations with seven different individual base models to identify the optimal ensemble configuration. Each base model was initialized with ImageNet pre-trained weights, with top classification layers replaced by global average pooling. Secondly, the proposed ensemble approach of MobileNetV2, ResNet50, and Efficient-NetB0 achieved superior performance with accuracy: 96.14%, precision: 94.10%, recall: 96.92%, and F1-score: 94.97%, outperforming all individual models and alternative ensemble combinations. Thirdly, this study compared the experiment results with several existing studies related to pneumonia classification. Fourthly, this study validated the proposed model on an external NIH pediatric dataset (94.73% ac-curacy) without fine-tuning, demonstrating true clinical transportability beyond benchmark dataset performance.