Gliomas, graded as type-IV tumors, are linked to poor prognosis and low survival chances. An accurate survival prediction model aids in strategically planning patients’ treatments. We derive a robust feature set for accurate survival days (SD) prediction, including radiomics, location-based features, and age from the triplanar segmentation network. We study features’ global and local impact on SD using various post-hoc explainable AI (XAI) methods. However, post-hoc methods can produce different results for SD prediction, raising the question of interpretability. Therefore, we cross-evaluated the results and found these post-hoc XAI methods were consistent in their interpretations, indicating the robustness of the feature set. Additionally, we establish the biological significance of imaging features better to understand their impact on tumor behavior and patient outcomes. Our SD prediction regressor model outperforms current methods. BraTS2020 validation results showed improvements of 37.7% in accuracy, 16.85% in mean squared error, and 85.8% in Spearman’s rank correlation compared to the top-ranking model of the BraTS2020 challenge.