Credit card fraud has become to be a significant issue over the past few years, as many people rely on credit cards for the purchases. Technological advances and increase of online shopping also responsible for the rapid increase in credit card fraudulent activity, which has resulted in large financial losses. In order to solve this issue, an efficient fraud detection system has been designed in this study using attention mechanism and 1DCNN-LSTM network for the classification of fraudulent credit card transactions. There are many machine learning algorithms that are commonly used to automatically identify credit card fraud but sometimes they do not take into account fraudulent activity that may lead to false alarms. This study aims to determine the best way to recognise credit card fraud events. The main motive of this research to develop a model to predict the credit card frauds by utilising deep learning and the SMOTE oversampling techniques. For detecting fraud transactions, a recurrent neural network with Bi-LSTM and attention mechanism is proposed in this study, which is very effective for processing sequential data with complex vector dependencies. The experiments show that presented model obtain a higher performance with validation accuracy 99.98% and test accuracy 98.99% which proves its efficiency to classify fraudulent transactions in comparison to other state-of-the-art classifiers XGBoost, Random Forest, Naïve Bayes, SVM, CNN and CNN with attention. It would be worth to noting that the presented model has potential to reduce the financial losses globally by recognising the credit card frauds instances or scams.