Pneumonia Classification using MobileNetV3Small and EfficientNetB0 Architectures

Abstract

Pneumonia is a lung infection that causes inflammation in the alveoli and remains a major health concern in Indonesia due to its high mortality rate. This study focuses on improving the accuracy of automatic pneumonia detection from chest X-ray images by optimizing model parameters in lightweight deep learning architectures. We apply two efficient convolutional neural network models, MobileNetV3Small and EfficientNetB0, and utilize transfer learning and data augmentation on a dataset of 3,913 chest X-ray images that have been balanced using downsampling techniques. The experimental results show that both models achieve strong classification performance, with validation accuracy exceeding 95%. EfficientNetB0 consistently delivers the highest accuracy after data augmentation, while MobileNetV3Small demonstrates faster inference and is more suitable for real-time applications. These findings indicate that lightweight CNN models combined with transfer learning provide an effective balance between accuracy and computational efficiency.