Antifraudster: A Multi-Perspective Real-Time Fraud Detection Framework for Multi-Participant E-Commerce Transactions

Department of Information Science and Engineering Acharya Institute of Technology Banglore, India

This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Fraud detection in e-commerce has been widely studied, and there are several models already that look at buyers, sellers, and transactions from different angles. These models, however have some key problems: they don't provide real-time monitoring, they use fixed thresholds that don't change with business size, they often need a lot of computing power that small businesses can't handle. To address these issues, this paper introduces Antifraudster, a new framework that improves multi-perspective fraud detection with three main features:(i) real-time fraud detection which give instant alerts, (ii) adaptive thresholds that adjust the fraud sensitivity based on vendor size, and (iii) an explainable AI layer that provides a clear reason for fraud predictions. The framework is also designed to be lightweight and efficient which makes it work well for both big e-commerce platforms and also smaller businesses. Tests show that Antifraudster improves detection accuracy, reduces false alarms, and builds more trust among vendors compared to existing systems. The recall for fraud detection is a bit lower (~82%) than the overall system accuracy of 90% which indicates that a small number of frauds might not be recognized immediately. Since the precision is still high(~90%), there are not many false alarms for valid transactions. Recall will be improved in future research in future work using combination machine learning techniques, more transaction metadata, and complex feature engineering.