Brain surgery is a crucial treatment for brain tumors, but neurosurgeons face significant challenges in accurately delineating tumor boundaries to maximize resection while preserving healthy tissue. “Hyperspectral imaging” (HSI) has emerged as a powerful intraoperative tool for real-time tumor detection, offering a non-invasive approach to assist neurosurgical decision-making. In this study, we propose a novel machine learning-based HSI framework for brain tumor classification and delineation, leveraging spectral-spatial feature extraction and robust classification techniques. Our approach was evaluated on an in-vivo brain dataset consisting of 61 HS images from 34 patients, achieving an impressive Random Forest accuracy of 100% and a Decision Tree accuracy of 99.75%, with SVM and Naïve Bayes classifiers also demonstrating high performance (99.06% and 95.29%, respectively). Additionally, our system also got a score of 1.0 for Random Forest and 0.9966 for Decision Tree, emphasizing our model’s strength. The AUC values ranged up to 1.0, emphasizing method’s effectiveness in tumor detection. The results presented here constitute a concrete point of reference for intraoperative brain tumor classification, demonstrating that HSI is a highly accurate and efficient decision-support tool in neurosurgical procedures.