ARCHIVES

The manual diagnosis of pneumonia from chest X-ray images is a time-consuming and labor-intensive process that is highly dependent on radiologist expertise. Misdiagnosis or delayed diagnosis can lead to severe clinical complications, especially in pediatric and geriatric populations. To address this limitation, an automated, robust, and explainable AI-based pneumonia detection system is proposed in this paper. The system utilizes an advanced machine learning pipeline integrating EfficientNetB0 for high-dimensional feature extraction, a Support Vector Machine (SVM) equipped with a Radial Basis Function (RBF) kernel for optimal boundary classification, and Gradient-weighted Class Activation Mapping (Grad-CAM) for visual explainability. By leveraging transfer learning through the uniformly scaled EfficientNetB0 architecture, robust, highlevel features are extracted from complex pulmonary radiographs while mitigating the risk of overfitting on limited medical datasets. The extracted features are efficiently classified by the SVM, achieving superior diagnostic performance including 92.5% accuracy, 95.1% sensitivity, and an F1-score of 91.8%. Furthermore, Grad-CAM heatmaps are overlaid on the original X-rays to highlight suspicious localized regions of opacification, substantially enhancing the interpretability of the model’s predictions. The entire pipeline is deployed through a robust, userfriendly Streamlit web dashboard, capable of generating comprehensive PDF medical reports containing the diagnosis, model confidence, and visual heatmaps. This transparency fosters trust among medical professionals and presents a highly viable, costeffective auxiliary tool for rapid clinical screening in diagnostic workflows.