Predicting mortality in patients with acute decompensated heart failure remains difficult for non-specialists. In addition, the influence of various heart failure complications on mortality has not been sufficiently confirmed. The purpose of this research is to assess the possibility of predicting the mortality risk in patients with acute decompensated heart failure after discharge using deep learning based on a registry of Japanese hospitalized patients with high rates of comorbid atrial fibrillation, chronic kidney disease, and anemia. We randomly divided data from fifteen clinical characteristics in 1,012 hospitalized patients into training and validation datasets. Next, we introduced the datasets into a prediction model using an automated-deep learning algorithm (Prediction One). Our deep learning-based model demonstrated a high ability to predict mortality risk (c-statistics = 0.75, sensitivity = 0.607, and 1 − specificity = 0.192). Prediction accuracy can be improved by appropriately incorporating input variables such as the brain-type natriuretic peptide level, red blood cell count, left ventricular ejection fraction, number of administered medications, length of hospitalization, and Nohria–Stevenson classification stage. We demonstrated that our deep learning model based on multiple clinical characteristics is useful for predicting the mortality risk in hospitalized patients with heart failure. In particular, we showed that our model including brain-type natriuretic peptide is effective for predicting the acute decompensated heart failure mortality risk.