Brain tumors are the fifth most prevalent type of cancer, impacting around 90,000 individuals each year and posing significant health risks due to their potential for severe neurological effects. Traditional diagnostic techniques often suffer from limited visual clarity, which can hinder accurate interpretation and delay critical treatment decisions. To address these limitations, this study presents an advanced deep learning model that integrates ResNet101, InceptionV3, DenseNet201 and VGG16 architectures for classifying brain tumors into four categories: pituitary, meningioma, glioma and no tumor. The model was trained and validated on a dataset containing 7,022 MRI images, with hyperparameter tuning to optimize classification accuracy. Additionally, Gradient-weighted Class Activation Mapping (Grad-CAM) was used to produce interpretable heatmaps, highlighting significant areas in MRI scans that influence classification. This level of interpretability enhances clinical trust, allowing the model’s focus to align with medically relevant regions. The combined model demonstrated high performance, achieving 99.2% accuracy, 98.8% precision, 98.7% recall and a 99.1% F1-score, with an AUC of 1.00 across all categories, indicating strong diagnostic capability. The Grad-CAM visualizations provided valuable insights for tumor classes, identifying critical regions in MRI images and supporting improved interpretability. This framework represents a significant advancement in brain tumor classification, streamlining the diagnostic process, reducing dependence on manual analysis and promoting more effective, accurate and sustainable healthcare solutions.