Credit Card Fraud (CCF) is a serious challenge facing credit card holder and the credit card delivering companies in the past decades. There are two levels CCF are performed, the transaction level frauds and application level frauds. This paper focuses on the application level of CCF detection using Genetic Algorithm (GA) as a feature selection technique. The GA feature selection technique is in two phases, the first phase is designated as the first priority features where eight (8) attributes were selected as the fittest attributes. At second stage which is referred to as the second priority features where another set of eight (8) attributes were considered and selected. The Naïve Bayes (NB), Random Forest (RF) and Support Vector Machine (SVM) supervised machine learning techniques were used for the detection of CCF on German credit card dataset which is an imbalance dataset. The experimental findings of the proposed model revealed that the first priority features are the most important features. Also, the obtained results showed that the RF algorithm outperformed NB and SVM in terms of accuracy, fraud detection rate and precision.