Brain tumors impose substantial global health burdens, with gliomas constituting approximately 25 % of all adult cases. Accurate grade classification between lower-grade glioma and glioblastoma multiforme is critical for treatment decisions, given their drastically different prognoses (90 % versus 5.1 % long-term survival). Traditional histopathological grading requires invasive biopsies and suffers from inter-observer variability, while existing machine learning approaches face limitations including reliance on complex preprocessing, synthetic data augmentation, lack of systematic feature selection, and singlemodel optimization that hinder clinical deployment. This study presents an AI-enabled framework for brain tumor glioma grade classification using clinical and mutation-based features from the UCI Machine Learning Repository containing 832 patient instances with 24 clinical and genetic attributes. We employ Mutual Information-based feature selection to identify the 10 most discriminative features, including Primary Diagnosis, IDH1, Age, PTEN, RB1, EGFR, ATRX, CIC, FUBP1, and IDH2. Ten diverse machine learning algorithms were evaluated through 5fold cross-validation, with top performers undergoing systematic GridSearchCV hyperparameter optimization. A strategic soft voting ensemble combining optimized XGBoost and CatBoost achieved 99.43 % accuracy with exceptional stability (±0.0012), alongside 99.51 % precision, 99.36 % recall, and ROC-AUC of 0.99. The confusion matrix demonstrated near-perfect classification with only 2 misclassifications among 257 test samples. This computationally efficient framework eliminates preprocessing complexity while maintaining interpretability through biologically relevant features, offering a practical tool for clinical brain tumor diagnosis and personalized treatment planning that significantly outperforms existing methods.