Brain tumors continue to be the most pressing health challenges, and accurate and timely diagnosis is crucial for a positive clinical outcome. Deep learning methods, particularly convolutional neural networks (CNNs) for detecting and classifying brain imaging data, have been shown to outperform more common imaging modalities like X-rays. However, these models are limited to the availability of MRI data which is relatively sparse. In this study, two benchmark datasets were employed, one with 253 images for binary classification (tumor, non-tumor) and a larger dataset with 3264 MRI images for multiclass classification to identify various tumor types. Various data augmentation and preprocessing techniques are employed improving the image quality and feature extraction. This approach achieved a classification accuracy 96.10% in binary classification and 91.24% in multiclass classification. The experimental results indicate that integration of CNN along with improved preprocessing methods provides better accuracy and speed for the identification of brain tumor. This approach has the potential for clinical applications, providing more rapid and robust diagnostic assistance.