The use of credit cards is growing in popularity and has become a billion-dollar industry. However, this has also led to an increase in fraud which has resulted in huge losses in many financial institutions. A lot of research goes into figuring out and forecasting fraudulent activities. Technology is also growing as fast as these fraudsters are. We employ various machine learning models to detect fraudulent activities in credit card fraud analysis problems. The problem of dealing with imbalanced datasets is perhaps one of the biggest in credit card fraud analysis. It was quite difficult to achieve high recall values which are very important in case of fraud detection. To address this challenge, our study recommends a model that employs hyperparameter-optimized XGBoost and a loss function that enhances the recall and AUC with balanced precision. In addition to outperforming conventional state-of-the-art models, our suggested strategy also performs better than the approach published in [1], which simply uses XGBoost and the method described in [2], which combines XGBoost, CatBoost, and LightGBM.