Pneumonia remains a leading cause of morbidity and mortality globally, disproportionately affecting vulnerable groups such as young children and the elderly. Timely and accurate detection of pneumonia is critical for prompt treatment and reducing mortality risks. This study aims to improve pneumonia detection performance by employing data augmentation techniques in deep learning models. The authors address challenges posed by insufficient and imbalanced medical image datasets, which often hinder model accuracy. Through strategies like rotation, zoom, translation, flip, and Zero-phase Component Analysis (ZCA) whitening, the authors proposed an augmented dataset from existing chest X-ray images. By using the augmented dataset, an intensive analysis for both VGG16 and conventional CNN has been conducted. The results show a big improvement in performance from using data augmentation, with VGG16 achieving 92.95% accuracy, 93.03% precision, and 95.9% recall, while the conventional CNN reached 86.86% accuracy, 85.98% precision, and 94.36% recall. Specifically, the efficiency of the augmented dataset has been verified, and the VGG16 model using augmented data achieved impressive metrics: an accuracy of 92.95%, a precision of 93.03%, and a recall of 95.9%. In contrast, the VGG16 model using the original dataset without data augmentation achieved only 70.03% accuracy, 67.59% precision, and 100% recall. The results demonstrated that the proposed augmentation method outperformed the original one. These findings underscore the potential of data augmentation in enhancing deep learning model performance, proposing a valuable approach for future clinical applications in medical diagnosis.