Credit card fraud is a big problem for banks and their consumers, and it costs a lot of money around the world. The data is quite unbalanced, making it difficult to identify fake transactions, which make up a very small fraction of all transactions. This study examines the utilization of machine learning methodologies for fraud detection, employing a publicly accessible dataset of credit card transactions conducted by European cardholders in September 2013. The dataset contains 284,807 transactions from two days, yet just 492 of these (0.172%) were found to be fraudulent. Principal Component Analysis (PCA) has changed all of the input variables except for the transaction amount and time to keep them private. In this work, we explore the difficulties of finding credit card fraud and try to look at the newest improvements in fraud detection methods, datasets, and evaluation standards. We list and assess the pros and cons of different ways to identify fraud. This study introduces a viable and reproducible methodology for detecting credit card fraud through supervised machine learning applied to a commonly utilized credit card dataset characterized by transactions over two days with significant imbalance. We talk about preprocessing, how to address very uneven class distributions, how to choose models, how to use evaluation metrics that work for rare-event detection, how to set up experiments, and what analysis we propose. We compare standard classifiers, including logistic regression, random forest, NB, and neural networks. We also talk about deployment issues and suggest future work.