Enhancing Glioma Grading Performance: A Comparative Study on Feature Selection Techniques and Ensemble Machine Learning

Glioma is a type of brain tumor that originates from glial cells. It can vary in severity and requires accurate grading for prognosis and treatment decisions. Manual grading methods are subjective and prone to variability, leading to machine learning algorithms to automate and improve the process. This research optimizes the machine learning models for glioma grading using three feature selection techniques: Genetic Algorithm-Logistic Regression (GA-LR), Boruta, and CatBoost Feature Importance (CFI). The study compares and analyzes these techniques’ impact on the ensemble models’ performance. The results show that the ensemble model SVM + RF + AdaBoost with CFI feature selection achieved the highest F1-score, i.e., 0.860. The CFI feature selection technique, which can identify relevant feature subsets, demonstrated better performance for glioma grading when compared to the previous study’s results. Therefore, this study concludes that CFI contributes to forming an optimal ensemble model, improving the accuracy of glioma grade classification.