The use of electronic payment methods for both in-store and online sales is becoming increasingly common among businesses worldwide. As the use of credit cards for online purchases has grown, so has the prevalence of suspicious conduct including internet fraud and payment defaults, the latter of which can result in substantial financial losses. Researchers investigated various machine learning classifiers that could detect anomalies in credit card transaction data to resolve this issue. Overlapping class samples and an uneven distribution of classes make it hard to spot outliers in this data. It is possible that general learning algorithms will prioritize samples from the majority class data from minority classes due to the low detection probability of abnormalities in minority class samples. To improve the rates of detecting credit card abnormalities, our proposed Credit Card Fraud Detection (CCFD) model uses numerous machine learning methods. Our solution to data imbalance and overfitting is the k-fold cross-validation procedure and the stratified sampling technique. Improving model performance and lowering the risk of overfitting are both achieved by working with optimized selected features instead of all features. A Mathew correlation coefficient (MCC) score of 0.958 and an improvement to 0.98 following optimization with Ejaya show that the Isolation Forest classifier outperformed the other classifiers on the German credit card dataset.