Enhancing Fraud Detection in Credit Card Transactions: A Comparative Study of Machine Learning Models

Abstract

The convenience and efficiency of credit cards for online transactions have led to their widespread adoption. However, this increased usage has also heightened the risk of credit card misuse and fraud. Such fraudulent activities result in substantial financial losses for both cardholders and financial institutions. The primary objective of this research study is to detect fraudulent activities using machine learning techniques, which offer promising avenues for accurate identification. However, the challenge lies in developing robust models capable of effectively distinguishing between genuine and fraudulent transactions. This study addresses this gap by proposing a systematic methodology for fraudulent transaction detection using machine learning (ML) algorithms. Three datasets are used, including two balanced datasets and one unbalanced dataset, as fraudulent transactions are typically far less frequent than genuine ones, ensuring the model’s generalizability. Through preprocessing techniques such as handling missing values and reducing dimensionality, combined with the implementation of machine learning models (Support Vector Machine (SVM), Decision Tree (DT)), ensemble learning (EL) models (Random Forest (RF)), Extreme Gradient Boosting (XGBoost), and Adaptive Boosting (Adaboost), and deep learning models (Multilayer Perceptron (MLP), Artificial Neural Network (ANN), the study explores various models’ effectiveness in capturing complex relationships within transactional data. The results demonstrate significant success, DT and RF, and MLP models achieving high accuracy of 0.99 underscoring their potential for accurate fraud detection in financial transactions. In the case of the unbalanced dataset, the RF model also performed effectively, achieving an accuracy of 0.97, with improved precision and recall for fraud detection.