Enhancing Brain Tumor Classification Using CNN: Impact of Data Augmentation on Model Generalization and Performance in Federated Learning

This paper presents an empirical evaluation of convolutional neural networks (CNNs) for brain tumor classification in a federated learning environment, comparing model performance with and without data augmentation. Federated learning ensures data privacy by keeping MRI datasets distributed among institutions while enabling decentralized model training across several compute nodes. Our experiments use CNNs trained on original and augmented MRI images, demonstrating significant improvements in test accuracy, precision, recall, and F1 score when data augmentation is applied, with the highest test accuracy reaching 95.51%. Augmented CNN models exhibit enhanced generalization across compute nodes, reducing training time and improving robustness in detecting brain tumors. The findings highlight how crucial data augmentation is to enhancing CNN models’ scalability and performance in federated learning, especially in medical imaging contexts where data variability is critical for accurate diagnosis and generalization across diverse datasets.