Objectives: To present the updated long-term outcomes form the Australia New Zealand Fontan Registry (ANZFR) and identify predictors of Fontan failure using machine learning. Methods: Patient data was obtained from the ANZFR. The primary composite end point was Fontan failure defined as any of death, transplant, Fontan takedown or conversion, protein losing enteropathy, plastic bronchitis or NYHA III/IV. Cox regression with L1 (lasso) regularization was used to identify associations between Fontan failure and 72 potential predictors using the machine learning ‘glmnet’ package in R. Parameter regularisation (shrinkage) was controlled by the lambda parameter with each lambda corresponding to a competing model. The best model was determined by having the highest Harrel C index, highest level of shrinkage, and 5-fold cross-validation. Lastly, predictors from the regularized regression with non-zero coefficients were included in a multivariable Cox regression model. Results: Out of 1859 patients in the Registry, over 48 years (1975-2023), 1617 (87%) patients were eligible for analysis. These included (n=221,13.7%) atriopulmonary, (n=275,17%) lateral tunnel and (n=1121, 69.3%) extracardiac conduit Fontans. Median age at Fontan was 4.6 years (IQR 3.68,5.81). Median follow up was 12 years (IQR 5.96,19.51). Freedom from Fontan failure at 10, 20 and 30 years was 90% (CI 88%, 91%), 78% (CI 75%, 81%) and 61% (CI 56%, 66%) respectively. 15 risk factors had non-zero coefficients which were then entered into multivariable Cox regression. Four of the 15 predictors had non-significant associations in the multivariable model and were removed. Conclusions: Freedom from Fontan failure at 30 years was 61%. Amongst 72 variables, machine learning demonstrated that AV valve repair, postoperative pleural effusion/chylothorax > 30 days, new onset arrhythmia and right ventricular dominance had the highest hazard ratios. Lateral tunnel and extra cardiac Fontan were associated with better outcomes.