Fast growth in technology affects all aspects of today’s life. Banking and financial payments also profit from this progress, but besides the facilities they serve, there are some disadvantages. Credit card fraud, known as the most prevalent fraud type, causes billions of dollars in loss for banking companies, financial industries, and their related customers every year. Although many solutions offer for preventing and eliminating credit card fraud based on up–to–date machine learning (ML) and, lately, deep learning (DL) methods to solve the problem, most of them do not strike a balance between speed and performance. Disinclination of financial industries to reveal their fraud dataset, which causes putting their reputation at stake, added more challenges. In this study, a prediction model for credit card fraud detection using Ensemble DL models by considering the above challenges to build a more confident and applicable model has been proposed. An Energy-based Restricted Boltzmann Machine (EB-RBM) and Extended Long Short–Term Memory (xLSTM) as base classifiers for developing a bootstrap max-voting ensemble model have been utilized. Final decision-making ensemble model considers %50 voting rights for RBM and LSTM classifiers equally, then normalize and aggregate the results to predict whether it is a fraudulent or genuine transaction. AUC–ROC, AUC–PR, precision, recall, F1–measure, confusion matrix, and elapsed time have been considered as evaluation metrics. The experimental results on the real-world European cardholder dataset reveal that the proposed ensemble model is more efficient in terms of the balance between speed and performance, among the recent models in the field.