World Journal of Engineering Research and Technology

Muntasir Hasan Kanchan and Muhammad Masud Tarek*

ABSTRACT

Detecting fraud in credit card purchases is perhaps one of the better testbeds for computational intelligence algorithms. Indeed, there are a variety of significant problems in this issue: definition drift (evolving consumer preferences and shifting tactics over time), class imbalance (actual transactions far beyond fraud), and latency verification (only a limited number of transactions are tracked in good time by the investigators). Accurate identification and avoidance of fraud are essential to protect financial institutions and individuals. The credit card fraud monitoring system was used to track fraudulent practices that was implemented. In this work, we use human explainable AI technique Local Interpretable Model-agnostic Explanations (LIME) to model the sequence of transactions in managing credit card transactions and show how it can be used to detect fraud. Decision Tree, K-Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM), and XGBoost, with three performance measures (accuracy, f1-score, and confusion matrix) required to demonstrate the classification prediction's effectiveness. As a result, it is vital to evaluate if a model generates a specific prediction. We eventually train an interpretable model called LIME for the sample based on its neighbors, this cardholder?s activity patterns, and the associated cross features. Compared to the five classifiers, KNN gives better results from accuracy and f1-score to identify fraud.

[Full Text Article] [Download Certificate]