Convolutional Neural Networks (CNNs) play a crucial role in medical image analysis, particularly in the detection of pneumonia, by facilitating accurate diagnosis and treatment planning. This paper presents a robust approach to pneumonia classification using the publicly available RSNA Pneumonia Detection Challenge dataset. The proposed method employs an ensemble-based CNN framework that integrates multiple pre-trained architectures, specifically MobileNetV2, InceptionV3, and ResNet152V2, through a combination of stacking techniques and soft voting. This ensemble strategy aims to leverage the individual strengths of each model for improved performance. All models are trained and evaluated exclusively on the RSNA dataset, and their effectiveness is assessed using AUC, accuracy, and F1-score metrics. The results demonstrate that the ensemble consistently outperforms individual models in terms of prediction stability and classification accuracy. This work highlights the potential of deep model aggregation techniques to enhance diagnostic reliability in clinical pneumonia screening systems.