The rapid growth in digital payments has led to a rise in credit card scams around the world, causing consumers, stores and banks to lose money. This paper addresses this problem by building and testing a machine learning system to quickly and accurately detect credit card scams. This study uses a set of 555, 719 payments from Kaggle with 2, 145 scams. After using SMOTE to prepare the data and correct unequal clusters, the study performs detailed data analysis. The analysis reveals key risk signals, such as 'shoppingₙet' and 'groceryₚause' spend peaking during late night (10: 00-11: 59 pm) scams and affecting card users aged 35-40 the most. It also suggests that gender does not change the risk of fraud. The main part of this study is to use, adjust, and compare four machine learning methods: Artificial Neural Network (ANN), Decision Tree (DT), K-Nearest Neighbor (KNN), and XGBoost. The success of the model is checked with accuracy, precision, recall, F1-score and confusion matrix. The K-Nearest Neighbor (KNN) method works best, with a test accuracy of 98. 39%, precision of 96. 90%, recall of 99. 99% and F1-score of 98. 42% making it the best in distinguishing scams from real payments. The study reaches its goal by detecting risky payment habits and building a very accurate prediction model. It states that KNN is a good choice for live scam detection systems. It suggests banks use this model, add better security like twofactor authentication during risky times, and educate consumers. The study suggests incorporating data from multiple countries could improve the model's predictive power and generalizability.
Adedokun et al. (Tue,) studied this question.