Pneumonia is a global health challenge that continues to affect the lives of millions. Innovative, low cost and time-saving diagnostic approaches are needed to detect pneumonia and have a timely treatment for the same. This research introduces Deep Learning approaches, specifically working on Resnet-18 and Resnet-50 architectures to detect pneumonia using chest X-ray images. A standard and pivotal dataset, the Radiological Society of North America (RSNA) Pneumonia Detection Dataset has been used in this study. This dataset contains 26,684 X-ray images, which are then preprocessed using advanced techniques including but not limited to cleaning, and equal distribution of pneumonia and normal cases. The number of parameters in the architecture is an important indicator of the complexity of the model, with ResNet-18 featuring 11.2 million while Resnet-50 features 24.8 million parameters. Even having the higher complexity, Resnet-50 has a remarkable accuracy of 89.06% and recall of 86.23%, highlighting great performance in detecting pneumonia cases. On the other hand, Resnet-18 has an accuracy of 82.56% which showcases efficiency in a shorter duration of training. The conducted comparative analysis focuses on the strengths of both the architectures. ResNet-18 offers a balance of precision (91.43%) and recall (76.20%) with computational efficiency, while ResNet-50 excels in accuracy and recall. This research explores the important change by not only improving the accuracy of the diagnosis but also the easier understanding of the deep learning model.