Early detection of brain tumor is essential for cancer diagnosis, as it can significantly improve survival rates. Magnetic resonance imaging (MRI) images of brain tumors must be performed early for diagnosis. MRI image’s unparalleled image quality makes it a preferred tool for this purpose. In this work, we have modified the U-Net architecture to identify different types of brain tumors - Edema, Enhancing tumor, Non-enhancing tumor and Necrosis in MRI images. We assess our model's performance utilizing the benchmark dataset Brats 2020, with optimization conducted through ablation studies on layer architecture, activation functions, loss functions and hyper-parameters. To determine the evaluation of the model, performance metrics such as specificity, precision, accuracy, sensitivity, meanIoU and dice coefficient are used, achieving a test accuracy of 99.45% with the Adam optimizer and a learning rate of 0.001. Our proposed approach surpasses previous research, showcasing its capability for swiftly and precisely classifying brain tumors.