Pneumonia Disease Detection Using Deep Learning

Pneumonia remains a leading cause of morbidity and mortality globally, especially among children, elderly individuals, and immune compromised patients. Accurate and early detection is vital for effective treatment, yet traditional diagnostic approaches—primarily chest X-ray interpretation by radiologists—are prone to subjectivity and time constraints. To address this, the proposed project employs deep learning-based automation to enhance diagnostic precision and reduce dependency on manual radiological assessments. Leveraging convolutional neural networks (CNNs), the system aims to categorize chest X-ray images into either “pneumonia-positive” or “normal,” enabling early intervention and improving clinical outcomes. This study integrates and evaluates three state-of-the-art deep learning models—Xception, EfficientNetB4, and EfficientNetV2S—using a publicly available dataset compiled from four distinct Kaggle repositories. The combined dataset comprises over 18,000 labeled images and is preprocessed through data cleaning, augmentation, normalization, and class balancing to ensure robustness and model generalizability. Each model is trained and validated on stratified image splits, ensuring balanced learning and performance evaluation. Among the three, Xception outperformed the others, achieving a classification accuracy of 95%, while EfficientNetB4 and EfficientNetV2S followed with accuracies of 79% and 74%, respectively. The implementation harnesses Python as the primary programming language, with TensorFlow and Keras frameworks powering model development and training. Key preprocessing and analysis are performed using NumPy, Pandas, and Matplotlib, while Scikit-learn is employed for performance metrics and visualization. Jupyter Notebook serves as the interactive development environment, streamlining the workflow and supporting reproducible research. The dataset is processed and stored locally, though Google Colab and cloud GPUs such as NVIDIA Tesla T4 were optionally used for training acceleration. Optimization techniques including dropout regularization, data augmentation, and the Adam optimizer were applied to improve generalization and reduce over fitting. Overall, the system offers a scalable, efficient, and accurate solution for pneumonia detection in resource-constrained clinical settings. Its end-to-end deep learning pipeline—spanning data ingestion, model training, evaluation, and prediction—demonstrates the practical utility of AI in modern healthcare diagnostics. Moreover, the project emphasizes open-source development and cost-effective deployment strategies, making it highly accessible for global health initiatives. Future enhancements may include multi-disease classification, integration of patient metadata, and deployment on edge devices for real-time diagnosis. Key Words: Scikit-learn, Keras, Pneumonia, Chest X-ray, Normalization, Classification accuracy, Stratified image splits, Convolutional Neural Networks