Abstract :

The Rapid Growth Of Digital Transactions Has Significantly Increased The Risk Of Financial Fraud, Making Real-time Fraud Detection Systems Essential For Secure Payment Processing. Traditional Rule-based Systems Are Increasingly Ineffective Due To Their Inability To Adapt To Evolving Fraud Patterns. This Research Proposes An Adaptive Real-time Payment Fraud Detection System Leveraging Machine Learning Techniques, Specifically The Isolation Forest Algorithm, Combined With A Drift Detection Mechanism Using Adaptive Windowing (ADWIN). The Proposed System Is Designed To Identify Anomalous Transactions By Analyzing Behavioral And Contextual Features Such As Transaction Amount, Location Score, Device Trust Score, Transaction Frequency, And Temporal Patterns. The Isolation Forest Model Is Utilized Due To Its Efficiency In Detecting Anomalies In High-dimensional Datasets Without Requiring Labeled Fraud Data. This Unsupervised Learning Approach Makes The System Scalable And Suitable For Real-world Deployment Where Labeled Datasets Are Often Limited Or Imbalanced. To Enhance Adaptability, The System Integrates A Simplified ADWIN Algorithm That Continuously Monitors Incoming Transaction Streams To Detect Concept Drift. Concept Drift Refers To Changes In Data Distribution Over Time, Often Caused By Evolving Fraud Strategies. By Detecting Drift In Transaction Patterns, The System Can Dynamically Respond To Changes And Maintain High Detection Accuracy.The Architecture Includes An Integrated Preprocessing Pipeline That Performs Feature Engineering, Such As Extracting Time-based Attributes And Scaling Numerical Features Using StandardScaler. The Processed Data Is Then Fed Into The Trained Model For Anomaly Detection. The System Outputs A Fraud Classification Along With A Confidence Score, Enabling Decision-makers To Assess The Severity Of Each Flagged Transaction. Additionally, The System Is Implemented Using A Django-based Web Interface, Allowing Users To Input Transaction Details And Receive Real-time Fraud Analysis. This Enhances Usability And Demonstrates The Practical Applicability Of The Solution. Experimental Results Show That The System Effectively Identifies Suspicious Transactions While Maintaining Low False-positive Rates. The Combination Of Isolation Forest And ADWIN Ensures Both Accuracy And Adaptability, Making The Proposed Model Suitable For Dynamic Financial Environments. In Conclusion, This Research Contributes A Robust, Scalable, And Adaptive Fraud Detection Framework Capable Of Handling Real-time Data Streams And Evolving Fraud Patterns. Future Work May Involve Integrating Deep Learning Models And Expanding Feature Sets For Improved Performance.

Published:

07-4-2026

Issue:

Vol. 26 No. 4 (2026)

Page Nos:

1563-1570

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite