The last couple of years have witnessed a limitation in personal mobility due to restrictions and constrains caused by the COVID-19 pandemic. Leading to an increase in online purchase for many goods and services, and pushing dependency on credit card use to a new record high. Credit card fraud presents a crucial challenge for both consumers and financial institutions. Therefore, there is a real need to develop a strong system to prevent fraudulent credit card transactions. Machine learning can provide solutions to avoid mostly all fraudulent credit card transactions. This paper evaluates 11 machine-learning models and presents a deep learning model for credit card fraud detection using a two-phase evaluation process. Each model uses the same real world credit card dataset. The two phase study first testes the eleven machine-learning algorithms and nominates the four best algorithms to reevaluate in the second phase. Machine learning algorithms and synthetic minority oversampling technique (SMOTE) are used to create the proposed model. The first phase indicates that K-neighbors-classifier (KNN), Support-vector-machine (SVM), and Gaussian-naïve-bayes-classifier (GNB) are the best among the eleven models. The second phase result reveals that the proposed model with SMOTE outperforms the nominated models with an AUC value of 100.00%, an F1-Score value of 99.00%, Precision value of 99.90% and a Recall value of 99.840%