The classification of brain tumors through Magnetic Resonance Imaging (MRI) is of paramount importance for the facilitation of early diagnosis and the formulation of treatment strategies; however, manual interpretation continues to be labor-intensive and susceptible to subjective bias. This manuscript proposes a hybrid deep learning framework that integrates transfer learning with attention mechanisms and traditional machine learning methodologies to enhance the detection of brain tumors. In particular, the Convolutional Block Attention Module (CBAM) is integrated into pretrained Convolutional Neural Networks—including ResNet50, DenseNet121, MobileNetV2, InceptionV3, and InceptionResNetV2—by positioning the attention module in advance of the global average pooling layer. This integration serves to augment the spatial and channel-wise emphasis on features pertinent to tumors. Subsequently, deep features gleaned from the attention-enhanced networks are subjected to classification employing a Support Vector Machine (SVM), chosen for its resilience when applied to small and imbalanced datasets. The proposed architecture is evaluated using three public brain MRI datasets. Research findings demonstrate that incorporating CBAM leads to a notable enhancement in classification accuracy across various architectural models, with DenseNet121 delivering the most consistent performance. Furthermore, the CBAM + SVM combination outperforms conventional Softmax classifiers, demonstrating enhanced generalizability and diagnostic reliability. This work highlights the efficacy of attention-guided transfer learning models in medical imaging and supports their integration into practical clinical decision-making systems.