TITLE:
Supervised Machine Learning Predictive Modeling of Survival Times of Malignant Glioma Patients
AUTHORS:
Erasmus Tetteh-Bator, Lohuwa Mamudu, Seth K. Laryea, Chris P. Tsokos
KEYWORDS:
Gliomas, Supervised Machine Learning, Loss Function, Generalized Pareto Distribution, Artificial Neural Network
JOURNAL NAME:
Journal of Data Analysis and Information Processing,
Vol.14 No.1,
February
5,
2026
ABSTRACT: Gliomas are the most cancerous tumors arising from the brain, accounting for approximately 78% of all primary brain tumors. The progression of malignant gliomas significantly reduces both survival duration and quality of life, with long-term survival achieved by fewer than 5% of patients, with a significant portion facing a grim outlook within 1.5 years, despite standard treatment. To improve survival prediction and support clinical decision-making, this study analyzes real-world data from the SEER (Surveillance, Epidemiology, and End Results) database of the National Cancer Institute (NCI). In this study, we identified a well-defined probability distribution that characterizes the survival times of patients with Malignant Glioma (3-Parameter Generalized Pareto Probability Distribution) and derived its cumulative distribution function (CDF) and survival function, providing estimates of the probability of patient survival. This information proves invaluable to healthcare professionals and the families of patients, addressing questions raised about patients, such as the patient’s survival probability given survival beyond a specified time, or the patient’s risk or intensity of death (hazard) at time t after surviving to that point. This study developed an Artificial Neural Network (ANN)-based survival analysis model, with a binary classifier predicting patient survival less than 12 months or beyond, and a multiclass model predicting survival within the first 12 months, between 12 and 24 months, or beyond 24 months. Model performance is evaluated using multiple metrics and compared with other machine learning algorithm approaches.