Credit card frauds can result in substantial financial losses, particularly when fraudulent transactions have large values. Thus, it is essential to detect fraudulent transactions prior to their authorization by card issuers. Most conventional fraud detection systems are based on machine learning models. Recent studies explored utilizing deep learning (DL) models to detect fraudulent transactions efficiently. However, such studies depend merely on a single DL model. In this paper, we present various deep learning and ensemble methods for detecting credit card fraudulent transactions. The main motivation behind this presented work is to contribute toward reducing both missed frauds and false alarms, where our contribution in this work lies specifically in combining the resulting scores of three different distinct DL models, namely, convolutional neural networks (CNN), autoencoders (AE), and recurrent neural networks (RNN). Experiments on a public credit-card dataset demonstrated that, for the single DL-based models, AE has the best validation accuracy (93.4%) compared to CNN (91.4%) and RNN (91.8%). For the ensemble results, the validation accuracy (94.9%) was superior to all the three implemented DL-based models.