A Systematic Survey of Machine Learning Methodsfor Fraud Identification in Credit Cards

Abstract

Credit card fraud is still a huge issue for the
financial industry, and over the years, it has caused the industry
to lose billions of dollars across the globe. The once-efficient
traditional rule-based systems are now almost entirely obsolete.
The primary focus of this survey is the shift from conventional
rule-based systems to astute, data-driven solutions for credit
card fraud detection. It even goes so far as to thoroughly
examine ML-based credit card fraud detection (CCFD). Along
with the difficulties of class imbalance, idea drift, and
verification delay, the study offers a comprehensive analysis of
the concept and different kinds of credit card transactions. It
also discloses publicly available datasets, including Kaggle,
IEEE-CIS, and PaySim. This research highlights the
contributions of preprocessing methods, such as data cleaning,
normalization, and PCA, to enhancing data quality and also
discusses ethical dilemmas related to transparency, bias
reduction, and informed consent. Besides that, the paper
discusses various supervised learning methods (Logistic
Regression, Decision Tree, Naïve Bayes) as well as different
unsupervised learning approaches (K-means, DBSCAN,
Autoencoders,  One-Class SVM) that can be applied to identify
fraudulent transactions. The survey reveals that ML models are
responsible for improving the accuracy, flexibility, and speed of
detection, which, in turn, can lead to the establishment of safer,
more trustworthy financial systems.