Ischemic stroke is characterized by the necrosis of brain tissue due to the blockage of blood vessels that carry blood to nourish the brain; it is a growing concern in a world with an increasing elderly population. Currently, attempts are being made to identify potential prognostic markers of ischemic stroke in the elderly, estimated pulse wave velocity (ePWV) being one such prognostic marker. Heightened ePWV values are associated with increased mortality. This study tries to characterize the ability of machine-learning models to predict long-term survival after ischemic stroke, based on the ePWV values. The study uses data from the publicly-available Medical Information Mart for Intensive Care-IV database, which itself contains data from adults admitted to the Beth Israel Deaconness Medical Center between 2008 and 2019. ePWV values were measured, and other lab parameters, vital signs, comorbidities, demographic information, and clinical severity scores that could potentially impact ePWV values were also measured. To allow for comparison between the control and intervention groups, Kaplan-Meier survival analysis and multivariable Cox proportional hazards regression were used to adjust for confounding variables. Four machine-learning algorithms - Logistic Regression, Random Forest, XGBoost, Naive Bayes - were also generated to capture any nonlinear relationships between variables and predictive outcomes. Overall, 1337 individuals' data was analyzed in this study. The mean ePWV value was found to be 9.3. Subjects were allocated into one of four quartiles (Q1: 3.9–7.6; Q2: 7.6–9.3; Q3: 9.3–11.3; Q4: 11.3–16.0) depending on ePWV values. Those in quartile 4 had higher ages, higher systolic and diastolic blood pressure, higher respiratory rate, higher heart rate, higher white blood cell count, higher levels of blood urea nitrogen, a higher incidence of comorbidities such as hypertension, heart failure, renal disease, and mild liver diseases, and increased severity of illness. Kaplan-Meier survival analysis showed that those with higher ePWV values had a risk of mortality after 30 days, 90 days, and 1 year. More specifically, those in quartile 4 had a significantly higher mortality risk compared to those in quartile 1, especially with regards to 30-day mortality. RCS analysis of the data showed a linear relationship between ePWV values and mortality risk after adjusting for the previously-mentioned lab parameters, vital signs, comorbidities, demographic information, and clinical severity scores. Among the 30-day predictive models, the Logistic Regression model was best with an area under curve (AUC) value of 0.829, while for the 90-day and 1-year predictive models XGBoost was the best with an AUC value of 0.845. The Logistic Regression, Random Forest, and XGBoost calibration curves all followed the reference curve quite closely, signifying their predictive accuracy. Shapley Additive exPlanations analysis was used on the Logistic Regression model to identify and rank the most influential measures in predicting mortality - ePWV values came within the top 5. The Area Under Curve Receiver Operating Characteristic (AUROC) was used with 95% confidence intervals to determine the machine-learning model's ability to discriminate - it was found that the Logistic Regression Model had statistically significant superiority compared to other models.