Heart Failure (HF) is a major cause of morbidity and mortality in the US. With aging of the US population, the public health burden of HF is enormous. We aimed to develop an ensemble prediction model for 30-day mortality after discharge using machine learning. Using an electronic medical records (EMR) database, all patients with a non-elective HF admission over 10 years (January 2001 - December 2010) within the Montefiore Medical Center (MMC) health system, in the Bronx, New York, were included. We developed an ensemble model for 30-day mortality after discharge and employed discrimination, range of prediction, Brier index and explained variance as metrics in assessing model performance. A total of 7,516 patients were included. The discrimination achieved by the ensemble model was higher 0.83 (95% CI: 0.80 to 0.87) compared to the benchmark model 0.79 (95% CI: 0.75 to 0.84). The ensemble model also exhibited a better range of prediction as well as a favorable profile with respect to the other metrics employed. In conclusion, an ensemble machine learning approach exhibited an improvement in performance compared to the benchmark logistic model in predicting all-cause mortality among HF patients within 30-days of discharge. Machine learning is a promising alternative approach for risk profiling of HF patients, and it enhances individualized patient management.