TITLE:
Comparative Analysis of ML Models for Survival Prediction of Glioblastoma
AUTHORS:
Muna Awel, Dave Rushit, Samantha J. Katner, Mansi Bhavsar
KEYWORDS:
Glioblastoma Multiforme, Machine Learning, Survival Prediction, Multi-Omics, Methylation Biomarkers, GridSearchCV, SHAP, Interpretability, ROC Curve
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.1,
January
19,
2026
ABSTRACT: Glioblastoma multiforme (GBM) remains one of the most aggressive brain malignancies, with a median survival of less than 15 months. This study advances glioblastoma multiforme (GBM) survival prediction by developing a comprehensive machine learning (ML) pipeline that integrates four classifiers: Logistic Regression, Random Forest, XGBoost, and Support Vector Machine (SVM) on TCGA-derived multi-omics datasets. Rigorous preprocessing, including missing data assessment (MCAR test), multicollinearity checks, and feature selection, was followed by hyperparameter optimization using GridSearchCV and 10-fold cross-validation to enhance model performance and generalizability. Predictive performance was evaluated with AUC-ROC, precision-recall curves, and classification reports, while interpretability was assessed through SHAP (SHapley Additive exPlanations) analysis to identify the most influential features driving survival predictions. Random Forest achieved the highest predictive accuracy while maintaining strong interpretability, highlighting key drivers of GBM prognosis such as age, MGMT promoter methylation status, and specific gene expression signatures. Despite promising results that demonstrate ML’s ability to handle GBM heterogeneity, limitations include the relatively modest sample size and lack of external validation. Future work will incorporate independent cohorts for external validation, explore advanced ensemble and hybrid modeling strategies, and further optimize models to meet clinical requirements for both accuracy and transparent decision-making.