This study investigates the efficacy of an ensemble model in classifying stroke images, combining CNN (Convolutional Neural Network), EfficientNetB7, and DenseNet201 architectures. Utilizing a dataset of 2,501 black-and-white images from the Kaggle stroke dataset, the research addresses the challenges posed by limited data and explores data augmentation techniques to improve model performance. The ensemble model’s performance is compared against individual models such as MobileNetV2, EfficientNetB0, ResNet50, and DenseNet201. Results demonstrate that, while the ensemble model shows potential, its accuracy does not significantly exceed that of the top-performing standalone models, highlighting the need for larger datasets and more sophisticated ensemble techniques to enhance reliability. This work provides insights into the application of ensemble learning for stroke classification, paving the way for advancements in AI-driven stroke diagnostics.