HYBRID ATTENTION-DRIVEN MACHINE LEARNING FRAMEWORK FOR BRAIN TUMOR CLASSIFICATION FROM MRI AND CT IMAGES

Accurate classification of brain tumors plays a crucial role in early diagnosis, treatment planning, and survival prediction. Although Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) provide detailed anatomical information, automated interpretation remains challenging due to intensity variations, imaging noise, tumor heterogeneity, and limited labeled datasets. Conventional deep learning models often suffer from overfitting, suboptimal feature representation, and limited interpretability. In this paper proposes a Hybrid Attention-Driven Machine Learning Framework that integrates deep attention-based feature extraction with handcrafted descriptors and ensemble classification for robust brain tumor categorization. The model combines lightweight convolutional neural networks enhanced with spatial attention mechanisms and texture-based statistical features derived from Gray-Level Co-occurrence Matrices (GLCM). Feature relevance is optimized using an Improved Particle Swarm Optimization (IPSO) strategy prior to classification. An ensemble of Random Forest, Support Vector Machine, and Gradient Boosting classifiers produces the final decision using majority voting. To enhance clinical transparency, SHAP and Grad-CAM visualizations are incorporated for explainable predictions. Experimental validation demonstrates high classification accuracy with improved generalization and reduced false diagnosis rates across heterogeneous MRI datasets. The framework offers a reliable, interpretable, and computationally efficient solution for automated brain tumor diagnosis.