A Survey on Credit Card Fraud Detection Using Holdout Cross Validation and Stratified K-fold Cross-Validation

Abstract :

In the area of credit card fraud detection, the implementation of machine learning algorithms is paramount to ensure the security of financial transactions. In this survey paper, we leverage the power of two robust cross-validation techniques, holdout and stratified K-fold cross-validation to explore decision trees, random forests, isolation forests, and k-means clustering in fraud detection scenarios. Considering the actual application, Decision Trees serve as an interpretable baseline, allowing transparent visualization of decision paths and helping identify fraudulent patterns. Random Forests, an ensemble of Decision Trees, reduce overfitting and improve prediction accuracy by aggregating various decision-making processes. An Isolation Forest, designed for anomaly detection excels at isolating anomalous credit card transactions. The ability to efficiently identify outliers provides effective protection against fraud. K-means clustering, on the other hand, divides transactions into clusters and highlights potential anomalies in the data. This paper evaluates the performance of these algorithms using Holdout and Stratified K-fold cross-validation. Holdout validation allows us to easily split the data training and test sets, while Stratified K-fold crossvalidation ensures balanced class representation in each fold. This is important for fraud detection scenarios with imbalanced datasets. This survey report sheds light on the evolving landscape of credit card fraud detection by examining the use of these algorithms and cross-validation techniques to achieve high levels of accuracy and security in realworld finance.