For effective treatment and diagnosis planning of brain tumor early detection and classification of brain tumors is essential. In this study, we are going to evaluate the performance of the MobileNet CNN model for brain tumor classification, with emphasis on adjusting various training parameters in order to improve different metrics such as accuracy, precision, recall, and loss. To determine MobileNet's applicability, we evaluate its performance with multiple epochs and learning rates using metrics such as accuracy, precision, recall, and loss. The results showed that MobileNet performs best with 10–20 epochs and a learning rate of 0.01, achieving excellent accuracy (92.19%-93.3%) and minimal loss (0.2–0.28). However, when training epochs is extended to 50, all the performance indicators declined, indicating overfitting and reduced generalization. These data show that MobileNet is most effective with shorter training durations and lower learning rates, emphasizing the importance of careful parameter tweaking in deep learning applications for medical imaging. This research contributes to developing optimized CNN architectures for reliable and efficient brain tumor classification.