Credit Card Fraud Detection

Abstract

This thesis focuses on building a machine learning model for detecting fraud in credit card transactions using a data set consisting of 100,000 transactions with 92,785 non-fraudulent and 7,192 fraudulent instances containing information about each of the transactions along with their merchant group or demographics information. The paper tackles imbalanced data problems using Extra Trees, Random Forests, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Logistic Regression machine learning models. The ensemble methods, Trees and Random Performed had 96.87% highest accuracy with precision, recall, and balanced F1 score (Score). when it comes to classifying fraud transactions among all of the models, it is also noticed that KNN had a fairly good performance with an accuracy of 96.29%, being only slightly lower than the ensemble model results. The best- performing model was the SVM (94.32% detection rate accuracy) and the worst one, but a really good performer in terms of complexity (the simpler model had 91.04% accuracy) was logistic regression. This study showcases the power of ensemble methods in the management of imbalanced data, offering high accuracies without sacrificing balance in performance metrics. The ensemble of Extra Trees, Random Forest, and KNN provided a trade-off between effectiveness and reliability making it the best choice to deploy in credit card fraud detection systems due to consistent and stable performance.