Accurate and efficient classification of brain tumors by magnetic resonance imaging (MRI) scans is essential for clinical follow-up and treatment planning. However, in deep learning models, computational costs are often a significant barrier to practical application. This paper presents Feature-Aligned Knowledge Distillation with XAI (FAKD-XAI), a novel framework that classifies and rationally interprets brain tumors in an efficient manner. FAKD-XAI combines logit-level Knowledge Distillation with an adaptive intermediate feature-level distillation from ResNet-50 (Teacher Model) to a lightweight MobileNetV3-Large (Student Model) to facilitate learning between complex and simple models. Our alignment module featuring a 1 \(\times \) 1 convolution layer was able to overcome the architectural divergences of the student model and enabled the efficient use of stratified feature transfer at different levels of the hierarchy. FAKD-XAI integrates Local Interpretable Model-agnostic Explanations (LIME), which enhances the understanding of the workings behind model predictions, leading to promoting trust from the clinicians. FAKD-XAI achieved an accuracy of 99.47% on the Brain Tumor MRI dataset while maintaining high computational efficiency, with an average inference time of 5.25 ms per image. This makes it highly suitable for practical, clinical deployment. The use of Explainable AI (XAI) confirms that the model focuses on pertinent tumor areas, suggesting FAKD-XAI’s usefulness as a reliable diagnostic aid. All code is available on GitHub: https://github.com/borhanitrash/FAKD-XAI .