The use of credit cards has grown significantly as the globe moves quickly toward digitization and cashless transactions. Additionally, there have been more fraud-related operations, which costs financial institutions a great deal of money. As a result, we must examine and distinguish between fraudulent and legitimate transactions. In this study, we wanted to implement the comprehensive model training procedure from beginning to end. As a result, we will have the best possible model to categorize the transaction into normal and aberrant varieties. Machine learning methods have been used in the fight against credit card fraud; however, fraud detection systems have yet to prove particularly effective. The relatively breakthrough of deep learning has been used in various fields to address complex challenges. In this study, we investigate several models for machine learning to spot fraudulent credit card activity. We compare the results produced by each model as well as their performance. The SMOTE methodology yields the most successful outcomes. It has been argued that under-sampling the majority class (the normal class) could effectively improve a classifier's sensitivity to the minority class. This research shows that, compared to just undersampling the majority class, the procedure we chose for oversampling the lower (abnormal) standard and undersampling the upper (normal) standard can enhance the analyzer's conduct.