article · FUOYE Journal of Engineering and Technology
As electronic commerce expands, credit card fraud has increased, prompting the widespread adoption of artificial intelligence for fraud detection. However, standard machine learning models frequently lack explainability, which creates trust issues and conflicts with legal demands for transparent automated decisions. To address this, several machine learning classifiers, including Decision Tree, Logistic Regression, K-Nearest Neighbours, Multilayer Perceptron, and Gaussian Naive Bayes, were tested on a public credit card dataset. All models showed strong performance in identifying fraudulent transactions, with Gaussian Naive Bayes reaching perfect scores across all evaluated metrics. To improve transparency, two explainable artificial intelligence techniques, SHAP and LIME, were incorporated into the detection process. The integration showed that explainability tools improve user perception of how models reason, with SHAP providing a slight advantage over LIME in clarifying the factors behind fraud predictions.
Automated systems detecting financial fraud must be both accurate and understandable. When algorithms block legitimate transactions or miss fraud, customers and institutions need to understand why. Adding explainability to fraud detection algorithms helps financial organisations meet regulatory transparency mandates while building customer trust by revealing the key transaction features, such as amounts and dates, driving each automated decision.
This work targets automated fraud detection applications for financial institutions, payment processors, and e-commerce platforms seeking to comply with legal transparency standards. The research represents applied, laboratory-tested work evaluated on a public dataset. Moving towards real-world commercial use would require validation on live transaction streams, advanced visualisations, and optimised training processes as noted in the findings.
AI-generated from the published abstract. Always read the original work before citing.
The rapid growth of electronic commerce has led to a significant rise in credit card usage, making it the most popular payment method for online and offline transactions. However, this surge has also resulted in increased credit card fraud. While Artificial Intelligence (AI) methods have been widely applied to detect these frauds, a major challenge lies in their lack of inherent explainability. The transparent nature of many AI models raises trust concerns among users and fails to meet legal requirements for AI transparency. Explainability is crucial in credit card fraud detection as it clarifies why a prediction was made and highlights key influencing features, such as transaction amount and date. Despite its importance, explainability in AI-driven fraud detection remains underexplored. This research uses a publicly available credit card dataset to investigate making fraud detection systems more explainable. Machine Learning (ML) methods such as Decision Tree, Logistic Regression, K-Nearest Neighbors, Multilayer Perceptron, and Gaussian Naive Bayes were implemented and evaluated based on accuracy, recall, and F1 score. Decision_Tree_Classifier achieved 99.93% accuracy, with high precision (99.86%) and recall (100.00%), Logistic_Regression had 99.86% accuracy, with 99.73% precision and 100.00% recall, K-Nearest Neighbors (KNN) showed slightly lower accuracy at 98.10%, with a minor trade-off in precision 96.29% but perfect recall, MLPClassifier matched Logistic Regression with 99.86% accuracy, 99.73% precision, and 100.00% recall, GaussianNB achieved perfect scores across all metrics, with 100.00% accuracy, precision, recall, and F1 score. These results demonstrate that each model was highly effective in distinguishing fraudulent from non-fraudulent transactions, with GaussianNB showing exceptional performance. To enhance interpretability, two Explainable AI (XAI) methods— SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were applied to the ML results. Findings indicate that XAI explanations can improve users' perception of the system’s reasoning. Among the methods, SHAP demonstrated a slight advantage over LIME in terms of explainability. Future work should focus on advanced visualization techniques, improved data preprocessing, and optimizing the training process to provide deeper insights for users.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.4314/fuoyejet.v10i1.13
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
