Brain gliomas, including astrocytomas, oligodendrogliomas, and mixed gliomas, present diagnostic and prognosis challenges. This study leverages RNA-Seq data obtained from TCGA database and machine learning approaches to distinguish between these glioma subtypes and tumor grades. By combining differential expression analysis with feature selection methods such as minimum Redundancy Maximum Relevance (mRMR) and Random Forest, implemented via the KnowSeq package, we identified compact and informative gene expression signatures. These signatures demonstrated strong potential for improving the distinction between glioma types and grades, yielding promising results. The best binary classification model for glioma subtype achieved an average accuracy of 84.15% and an F1-score of 84.7% in cross-validation. Regarding tumor grading, the astrocytoma grade model achieved 97.43% accuracy and 96.29% F1-score, while the oligodendroglioma grade model reached over 97% accuracy and F1-score.