Transcatheter aortic valve replacement (TAVR) is a widely adopted minimally invasive procedure in high-risk surgical populations and is associated with a substantial rate of unplanned hospital readmissions. Traditional approaches to readmission risk assessment have predominantly relied on conventional statistical models, while the application of machine learning (ML) methods remains limited. The increasing availability of large-scale data derived from electronic health records (EHRs) creates new opportunities to investigate advanced ML approaches for outcome prediction in complex perioperative pathways. This study evaluates and compares multiple ML models for predicting unplanned hospital readmission within 30 days and 1 year following an index TAVR procedure and identifies key predictors associated with early and late readmissions. Data from 1043 patients treated between 2011 and 2019 within a multi-hospital health system in the United States were analyzed. Baseline patient characteristics were assessed, extensive feature engineering was performed, and ML models were trained and tested at both prediction horizons. In the final cohort, 254 features (out of 1865) were retained, including measures capturing social determinants of health such as the area deprivation index. Readmission occurred in 10.9% of patients within 30 days and in 32.7% within 1 year. Ridge logistic regression and support vector machine models achieved the highest performance for 30-day readmission prediction (AUC = 0.72; 95% CI: 0.62–0.82), while ridge logistic regression performed best for 1-year readmission prediction (AUC = 0.69; 95% CI: 0.61–0.77). Beyond procedure-specific findings, this work illustrates how the structure, completeness, and longitudinal nature of EHR data critically influence the performance and generalizability of AI-based prediction models. These considerations are particularly relevant for emergency and acute care surgical settings, where data fragmentation and time-critical decision-making pose major challenges to the implementation of clinically actionable AI tools.