Glioma Classification: A Comparative Analysis of Supervised and Ensemble Learning Methods

The accurate differentiation between higher-grade gliomas (HGGs) and lower-grade gliomas (LGGs) is critical for optimizing treatment strategies and improving patient outcomes. Despite the substantial research on glioma classification, most studies rely on conventional machine learning models or deep learning methods, often overlooking the potential of ensemble approaches. This study introduces a novel comparative analysis of various supervised learning models and ensemble learning strategies, specifically focusing on their application to glioma classification using clinical and molecular biomarkers. Utilizing three distinct datasets from The Cancer Genome Atlas (TCGA) and the Chinese Glioma Genome Atlas (CGGA), we trained and evaluated eight individual supervised machine learning models and nine ensemble models employing hard voting mechanisms. Notably, this study emphasizes the importance of ensemble techniques in achieving enhanced stability and robustness across datasets. The results demonstrate that the linear support vector machine (SVC) achieved the highest accuracy of 90.1% on TCGA 1, outperforming all other individual models. Ensemble learning strategies, particularly combinations including linear SVC, AdaBoost, k-nearest neighbors (KNN), and random forest, consistently delivered superior performance across all datasets. Importantly, our analysis highlights that ensemble models not only outperform individual models but also provide more reliable classifications, especially in the context of class imbalances inherent in medical datasets. This work contributes to the growing body of research by demonstrating the potential of ensemble learning techniques for glioma classification and underscores their value in clinical decision-making, marking a step forward in the integration of machine learning with clinical oncology.