Explainable AI-Driven Insights for Accurate and Transparent Glioma Grading

Since gliomas are tumors of the brain, they are normally classified into two major categories: Low-Grade Glioma and the highly malignant Glioblastoma Multiforme. Accurate glioma grade classification will be useful for prognosis and therapeutic decisions. We aim to develop a novel solution for classifying glioma grades by integrating classical machine learning methods with the latest advancements in deep learning. The proposed framework thus consists of an ensemble involving Random Forest, Decision Tree, Logistic Regression, K-Nearest Neighbors, AdaBoost, SVM, CatBoost, LightGBM, XGBoost, ANN, and CNN. Once the features are selected, the Random Forest algorithm picks the best features of the data set, wherein rest of the classifiers are then used for further modeling of the selected features. Such feature selection reduces dimensionality whilst allowing features that are considered important to be included in modeling, which then improves learning efficiency. Besides that, this work extends the behavioral variety of the analysis phase by adding Bernoulli Naive Bayes (BernoulliNB), Linear Discriminant Analysis (LDA), and the Passive Aggressive Classifier. To improve interpretability and render the model useful in the clinical setting, the study framework tries to incorporate Explainable AI (XAI) techniques, especially the SHapley Additive exPlanations (SHAP) algorithm.