Background: Radical cystectomy (RC) is a standard treatment for patients with muscle invasive and certain high-risk non-invasive bladder cancers. Existing clinical risk calculators based for RC on standard regression methods have demonstrated poor, inaccurate predictions of post-operative risks and adverse outcomes. We aimed to build a risk calculator using novel machine learning methods to predict complication rates and other outcomes based on clinical characteristics of patients undergoing RC. Methods: Patients who had an RC for bladder cancer between 2019 and 2021 were pulled from the National Surgical Quality Improvement Database. Potential predictors included clinically meaningful characteristics such as patient demographics, pre-operative labs, comorbidities, pre-operative bowel and antibiotic preparations, and prior treatment history, as well as cystectomy-specific characteristics such as diversion type and surgical approach. Outcomes included risk of any complication, infectious complication, mortality, and readmissions. Logistic regression, random forest, and support vector machine algorithms were used to build predictive models. Models were optimized assessed with area under the ROC curve (AUROC) and average precision (AP). Results: 8,936 patients were included in the sample dataset. Random forest classifiers outperformed other models for predicting risk of any complication (AUROC = 0.662; AP = 0.66), infectious complication (AUROC = 0.615; AP = 0.20), and mortality (AUROC = 0.603; AP = 0.03). The logistic regression model best predicted readmissions (AUROC = 0.587; AP = 0.26). Top predictors of complication risk included operation time, glomerular filtration rate, and type of procedure (robotic vs. open). Infectious complication was best predicted by wound classification and diversion type, mortality was best predicted by BMI and operation time, and readmissions was predicted most by pre-existing renal failure and emergent cases. Conclusions: Machine learning-based risk calculators were more effective in predicting morbidity and mortality after RC compared to more standard predictive algorithms. Machine learning algorithms therefore demonstrate greater predictive potential when given a wide range of clinical characteristics.