<p>Magnetic Resonance Imaging greatly improves the identification and categorization of brain tumors. Many related studies often struggle with some major issues, including data imbalance, limited datasets, and challenges in accurately detecting tumor regions in complex images. Many conventional deep learning models, such as DenseNet-201, show excellent results but also suffer from overfitting issues. This paper proposes the Dual Attention Enhanced Gradient-Guided DenseNet model to overcome all these limitations. The proposed model integrates the DenseNet-201, a dual attention mechanism, and Gradient-Weighted Class Activation Mapping. The spatial and channel attention mechanism improves the DenseNet-201 architecture by focusing on crucial Magnetic Resonance Imaging regions, and this improves the classification accuracy. Gradient-weighted class activation mapping facilitates an effective decision-making process by integrating an interpretation layer. In order to optimize the input images for further analysis, the proposed model integrates crucial preprocessing methods like greyscale conversion, Gaussian blur-based noise reduction, and image resizing. Furthermore, data augmentation techniques like rotation, flipping, and scaling are used to improve the dataset and successfully address issues with class imbalance. In this paper, four datasets are used to validate the performance of a proposed model. The experimental findings demonstrate that the proposed model surpasses all other existing models in terms of classification accuracy. The model achieved a 98.87% F1-score, 99.32% accuracy, 98.65% precision, 99.14% recall, and 98.71% specificity using the Brain Tumor-Magnetic Resonance Imaging dataset.</p>