AI Powered Fraud Detection In Online Payment

Ms. Vaishnavi Madhukar Mali^1*, Mr. Sumeet Sambhaji Ingole², Ms. Dipali C. Kothavale³

---

ABSTRACT

Web based payment systems are becoming targets of fraud cases and this demands strict, quick and sound detection systems. The paper will suggest an AI-based fraud detection system and combine the behavioral feature engineering approach with the class imbalance correction strategy and hybrid machine learning infrastructures. The system presented offers a joint Gradient Boosting classifier and a Graph Neural Network (GNN) to ensure both statistic pattern and relationship between users, equipment, and merchants can be obtained at transactional level and relationship levels, respectively. SMOTE and focal loss methods are used in order to combat extreme class disparity. Detection of concept drift is done based on drift-sensitive sliding-window retraining approach. Stratified 10-fold cross-validation shows that the suggested hybrid model yields a PR-AUC of 0.947 + 0.012, which is better than the standalone CNN and GNN baselines and lower falses rate. Also, explainability is added with the help of SHAP and GNN saliency analysis to give clear decision guidance to financial systems. The findings indicate that the given framework enhances the robustness in the cases of fraud detection and at the same time sustains the operation interpretability which is needed in the real world payment conditions.

Keywords: Online payment fraud detection, machine learning, graph neural networks, adaptive ensemble learning.

How to cite this article: Mali VM, Ingole SS, Kothavale DC. AI Powered Fraud Detection In Online Payment. Int J Drug Deliv Technol. 2026;16(15s): 306-317. DOI: 10.25258/ijddt.16.15s.38

Source of support: Nil.

Conflict of interest: None