Robust Brain Tumor MRI Classification Using Weighted Ensemble of Transfer-Learned CNNs with Explainable Predictions

This resource presents a comprehensive study on brain tumor classification using MRI images, emphasizing cross-dataset generalization and model interpretability. The work evaluates three transfer-learned convolutional neural networks - ResNet50, DenseNet121, and ConvNeXt-Tiny - and proposes a weighted ensemble strategy to improve robustness across datasets. Experiments are conducted on a public Kaggle brain tumor MRI dataset and a merged external dataset, with results demonstrating superior performance and reduced misclassification. Grad-CAM visualizations are included to provide interpretable insights into model predictions, highlighting clinically relevant regions. The submission includes the research paper, code for preprocessing, model training, ensemble implementation, and Grad-CAM generation, and optionally the cleaned and merged MRI dataset labeled with four tumor categories: glioma, meningioma, pituitary tumor, and no tumor. This resource enables reproducible research and provides a foundation for further development of robust, interpretable MRI-based brain tumor classification systems.