Abstract :

Online Payment Platforms Have Become A Crucial Component Of Modern Financial Services, But They Are Increasingly Targeted By Fraudulent Activity. Detecting Fraudulent Payments Is Difficult Because Of The Class Imbalance Between Genuine And Fraudulent Transactions: Fraud Forms A Small Minority Of Any Transaction Log, So A Classifier Trained On The Raw Distribution Learns To Favour The Majority Class And Misses Precisely The Cases It Was Built To Catch. This Project Presents A Machine-learning Fraud-detection System That Addresses This Imbalance Directly By Applying The Synthetic Minority Over-sampling Technique (SMOTE) Before Training, And It Quantifies The Benefit By Evaluating Each Classifier Both With And Without Balancing. The System Is Implemented As A Django Web Application In Python, With Scikitlearn For Modelling, Imbalanced-learn For SMOTE, NumPy And Pandas For Data Handling, Matplotlib For Visualisation, And MySQL For User Management. A Transaction Dataset Of 28,213 Records Labelled Fraud And Non-Fraud Is Normalised Using Standard Scaling, Encoded With A Label Encoder, And Split 80:20, Giving 22,570 Training And 5,643 Test Records; SMOTE Then Expands The Training Partition To 31,886 Records With Equal Class Representation. Random Forest And Naive Bayes Are Each Trained On Both The Original And The Balanced Training Sets And Evaluated On The Same Held-out Test Partition. Random Forest With SMOTE Performs Best, Reaching 99.344% Accuracy, 99.016% Precision, 99.371% Recall, And 99.192% F-score, Which Corresponds To 5,606 Of 5,643 Test Transactions Classified Correctly, Against 97.572% For Random Forest Without Balancing, 78.611% For Naive Bayes With SMOTE, And 77.884% For Naive Bayes Alone. The Trained Model Is Then Applied To Uploaded Test Data And Labels Each Record As A Fraud Or Normal Transaction. Overall, The Results Show That Class Balancing Improves Both Classifiers And That A Balanced Random Forest Provides Accurate, Interpretable, And Deployable Online Fraud Detection.

Published:

06-10-2026

Issue:

Vol. 26 No. 10 (2026)

Page Nos:

7 - 17

Section:

Articles

License:

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

How to Cite