Cerebral tumors are among the most deadly types of cancer in the world, and the need for early detection highlights the need of novel analytical approaches. Brain tumors classified by shape, size, surface, and area as revealed by magnetic resonance imaging (MRI) are the primary focus of this study. Applying deep learning algorithms like as Convolutional Neural Networks (CNNs), Xception, ResNets, and DenseNet201 on Audience 1 and Stage 2 datasets has led to significant improvements in characterisation accuracy. More research into the topic revealed that Xception had made significant progress, achieving a surprising near-perfect accuracy for both datasets, even though CNN's accuracy was 88% initially. This demonstrates the value of using a variety of deep learning architectures, with Xception emerging as an especially potent tool for accurate brain cancer classification. In addition to providing an opportunity to improve diagnostic accuracy, these results lend credence to further research into cutting-edge methods for dealing with the enormous challenge that brain malignancies produce.