Purpose: Compare and rank machine learning (ML) and traditional risk models for predicting postoperative cardiac complications after non-cardiac surgery using a Bayesian network-meta-analysis. Methodology: Cochrane Library, Embase and Medline (OVID) were searched to 30 June 2025. The primary outcome was model discrimination for predicting postoperative cardiac adverse events, assessed using the area under the receiver operating characteristic curve, with mean differences (MD) calculated relative to the Revised Cardiac Risk Index (RCRI). The risk of bias was assessed using the PROBAST+AI tool. As many studies evaluated multiple versions of each model type, the highest performing (“Best Version”) and lowest performing (“Worst Version”) results were analysed. Models were ranked using Surface Under the Cumulative Ranking (SUCRA) values. A sensitivity analysis included only low risk of bias studies. Results: Thirteen studies totalling 927,113 patients were included. ML approaches generally outperformed traditional risk scores. Automated machine learning (AML), ranked highest, demonstrating the greatest improvement in the Best Version analysis (MD 0.285, 95% CrI 0.166 to 0.405) and remained superior in the sensitivity analysis (0.300, 0.146 to 0.451). Gradient boosting models (GBMs) ranked second with strong performance across both analyses. The Gupta score outperformed RCRI in the Best Version analysis (0.165, 0.012 to 0.320). Between-study heterogeneity was low across all analyses. Only 38.5% of studies were at low risk of bias, and no model underwent external validation. Conclusion: Most ML models demonstrated better discrimination than conventional surgical risk tools, with AMLs and GBMs ranking highest. However, study quality, calibration reporting, and absence of external validation limit immediate clinical adoption. Prospective, multicentre surgical validations studies are required before routine perioperative implementation.