Publication Details
Abstract
Online transactions, e-commerce, and online banking have significantly contributed to the growth of financial fraud in the digital payment systems. The conventional rule-based fraud detection systems can be easily compromised to detect complex and evolving fraud patterns, which result in a late detection of fraud and greater losses incurred. To overcome the challenges, this paper will suggest an AI-based hybrid system of real-time fraud-detection in the financial transactions that integrates multiple machine learning methods to enhance its detection accuracy and reliability of the system. The proposed model combines anomaly detection and classification models in order to efficiently detect suspicious transactions with less false positives. This study is based on the Credit Card Fraud Detection Dataset 2023 that includes more than 550,000 anonymized credit card transactions by European cardholders. The dataset contains anonymized features (V1- V28), amount of transaction and a binary label showing whether a transaction is a fraud or a legitimate transaction. The dataset is subject to several preprocessing methods, such as data cleaning, feature scaling, and addressing the problem of imbalance between the classes with the help of oversampling. Preprocessing procedures are necessary so that the models will learn to identify patterns of both legitimate and fraudulent transactions. The hybrid framework as proposed by the authors involves the combination of several machine learning models to improve the performance of fraud detectors. The system can both learn the known fraud patterns and detect abnormal behavior of a transaction by incorporating anomaly detection mechanisms with classification algorithms. The framework will facilitate real-time detection, whereby the transaction data is analyzed rapidly to determine the existence of possible fraudulent transactions and identify them as they happen. Various metrics are used to gauge the performance of the proposed model and they include accuracy, precision, recall, F1-score and ROC-AUC. The experimental findings reveal that the hybrid method enhances the ability to detect fraud more than the traditional methods that use single models. The proposed framework will help to create more powerful and smarter fraud detection systems that could help financial organizations to minimize financial losses and enhance the security of transactions.