Glioblastoma, the most prevalent and aggressive primary brain tumor in adults, poses substantial clinical management challenges due to its high recurrence rate, poor prognosis, and inherent complexity. This study aims to enhance the prediction of overall survival in GBM patients treated with stereotactic radiosurgery by leveraging advanced image segmentation and machine learning techniques. We introduce a novel combination of Cat Swarm Optimization-assisted hybrid ResNet and U-Net architectures for precise tumor segmentation, and an ensemble of machine learning algorithms for robust survival prediction, advancing beyond traditional methods. Utilizing the BraTS2020 dataset, we accurately delineated three critical tumor regions: core, edema, and enhancing tumor, achieving remarkable metrics, including a segmentation accuracy of 99.2%, a loss of 0.023,a recall of 0.986, a mean intersection over union (IOU) of 0.991, a dice coefficient of 0.96, a precision of 0.991, a sensitivity of 0.991, and a specificity of 0.997. For survival prediction, various machine learning models were tested, with the Random Forest algorithm demonstrating superior performance in managing the complexities of the derived features from segmented images. The Ensembling approach for survival prediction further improved accuracy, achieving 60.01%. These findings suggest that integrating sophisticated segmentation with machine learning can significantly enhance survival predictions in Glioblastoma patients. This approach not only improves prognostic accuracy but also offers potential advancements in clinical management and personalized treatment strategies.