Exploring Activation Functions for Enhanced Brain Tumor Classification with Xception and Custom Dense Layers

Brain tumor detection is crucial for early diagnosis and treatment, improving patient survival rates, and reducing the risk of severe neurological damage. This paper presents a deep learning approach for classifying brain MRI images using the Xception model with custom dense layers. Transfer learning is employed to leverage pre-trained weights from the ImageNet dataset, enhancing the model's ability to extract complex features. Various activation functions, including ReLU, Mish, and Swish, are tested, with the ReLU-based 1024 dense -layer architecture delivering the best performance. Data augmentation techniques and optimization methods, such as the Adam optimizer and early stopping, are applied to ensure robust model generalization. The model achieves a high validation accuracy of 99.44%, demonstrating its potential in clinical diagnosis. The study also highlights the influence of various activation functions and architectural choices unon classification performance.