Brain tumors are among the most difficult clinical conditions, requiring accurate identification and localization to guide therapy and improve survival. Their variability in size, morphology, and location presents a problem for automated detection, highlighting the importance of explainable artificial intelligence models in fostering trust and reliability in practice. XRAI (eXplanations through Region Activation Integration) is a state-of-the-art explainability method that generates visualizations of the regions that affect a model's predictions, thus allowing clinicians to verify and interpret AI results efficiently. In this research, ResNet101V2, MobileNet, and InceptionV3 were tried out for brain tumor classification with emphasis on combining explainability with XRAI. MobileNet recorded the higher classification accuracy of 98%. In comparison, ResNet101V2 was much less accurate at 35%; nonetheless, it generated highly interpretable XRAI saliency maps and always circumscribed the tumor areas. This feature enables more effective tumor localization, thus satisfying the critical requirement of interpretable and reliable AI tools in medical imaging.