Heart rate variability (HRV) is an effective predictor of congestive heart failure (CHF). However, important challenges exist regarding the effective temporal feature extraction and efficient classification using high-dimensional HRV representations. To solve these challenges, an ensemble method for CHF detection using short-term HRV data and deep neural networks was proposed. In this paper, five opensource databases, the BIDMC CHF database (BIDMC-CHF), CHF RR interval database (CHF-RR), MITBIH normal sinus rhythm (NSR) database, fantasia database (ED), and NSR RR interval database (NSR-RR), were used. Additionally, three RR segment length types (N = 500, 1000, and 2000) were used to evaluate the proposed method. First, we extracted the expert features of RR intervals (RRIs) and then built a long short-term memory-convolutional neural network-based network to extract deep-learning (DL) features automatically. Finally, an ensemble classifier was used for CHF detection using the above features. With blindfold validation (three CHF subjects and three normal subjects), the proposed method achieved 99.85%, 99.41%, and 99.17% accuracy on N = 500, 1000, and 2000 length RRIs, respectively, using the BIDMC-CHF, NSR, and FD databases. With blindfold validation (six CHF subjects and six normal subjects), the proposed method achieved 83.84%, 87.54%, and 85.71% accuracy on N = 500, 1000, and 2000 length RRIs, respectively, using the NSR-RR and CHF-RR ndatabases. Based on feature ranking, the significant effectiveness provided by the DL features has been proven. The results have shown that the deep ensemble method can achieve reliable CHF detection using short-term heart rate signals and enable CHF detection through intelligent hardware.