Brain tumor is a critical medical condition where early and accurate diagnosis of the presence and type of tumor is essential for effective treatment. Manual diagnosis of brain tumor is a time-consuming process and are susceptible to human inaccuracies. The need for precise prediction highlights the importance of AI based solution. This research presents a comparative study of deep learning models for the detection of brain tumor utilizing MRI images. A dataset which contains over 30,000 labelled images has been used in the training process and assess four models such as CNN,ResNet18, MobileNetV2 and EficientNetB0. These models are assessed according to their classification accuracy and ROC-AUC which are performance metrics. Among the four models, ResNet18 achieves highest accuracy of 90.95%and with a ROC-AUC of 0.9589, followed by MobileNetV2 with accuracy pf 89.37% and ROC-AUC of 0.9688. The CNN model was evaluated with 86.72% accuracy with a ROC-AUC of 0.9278 whereas Efficient achieves 81.17% accuracy with a ROC-AUC of 0.9681. These findings represent that ResNet18 and MobileNetv2 are the most reliable models for the real world medical applications indicating the efficiency of deep learning approaches.