Smart Fraud Detection System for Shoprite Online Payments: A Random Forest Approach with Leakage-Controlled Evaluation and a Protocol for Collecting Verified Labels

Online grocery payments by card and mobile money are exposed to fraud, yet a retailer rarely holds the fraud-labelled data that supervised detection needs. This study presents a smart fraud detection system for the online payments of Shoprite, built as an academic case study with synthetic checkouts and public datasets; no Shoprite data were used. A transparent rule engine scores every order and sends suspicious ones to an administrator for review. A Random Forest model is consulted only when an order has the fields it was trained on; otherwise the system reports an explicit unsupported status and creates no score. Analysts record verified outcomes for later training. We tested the Random Forest and two baseline classifiers on two public datasets, 283,726 card transactions (473 frauds) and 235,370 web addresses (100,520 phishing), splitting before any fitting, grouping addresses by website and choosing thresholds without the test data. On held-out card data the Random Forest was the best model, with an area under the precision-recall curve of 0.800 (95% confidence interval 0.713 to 0.876), precision 0.926 and recall 0.789; oversampling before the split inflated the same model to 1.00. On web addresses a logistic model scored 0.999, but every legitimate address in the data was a secure site root starting with www and having no path; rewriting phishing addresses in the legitimate format cut its detection from 99.5% to 29.6%. A robustness gate now blocks such models, and a host-name model that replaced it detected 52% of phishing addresses at a 0.8% false-alarm rate. No labelled Shoprite fraud data exist yet, and the paper sets out how to collect them.

Authors

Publication Details

Journal
International Journal of Advanced Multidisciplinary Research and Studies
Published
2026-11-17
DOI
https://doi.org/10.62225/2583049x.2026.6.6.7173
Primary Topic
Imbalanced Data Classification Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Smart Fraud Detection System for Shoprite Online Payments: A Random Forest Approach with Leakage-Controlled Evaluation and a Protocol for Collecting Verified Labels

Gloria Ilunga Kasongo, Sikwa .
International Journal of Advanced Multidisciplinary Research and Studies
Imbalanced Data Classification Techniques
article

Smart Fraud Detection System for Shoprite Online Payments: A Random Forest Approach with Leakage-Controlled Evaluation and a Protocol for Collecting Verified Labels

Gloria Ilunga Kasongo, Sikwa .
article en

Abstract

Online grocery payments by card and mobile money are exposed to fraud, yet a retailer rarely holds the fraud-labelled data that supervised detection needs. This study presents a smart fraud detection system for the online payments of Shoprite, built as an academic case study with synthetic checkouts and public datasets; no Shoprite data were used. A transparent rule engine scores every order and sends suspicious ones to an administrator for review. A Random Forest model is consulted only when an order has the fields it was trained on; otherwise the system reports an explicit unsupported status and creates no score. Analysts record verified outcomes for later training. We tested the Random Forest and two baseline classifiers on two public datasets, 283,726 card transactions (473 frauds) and 235,370 web addresses (100,520 phishing), splitting before any fitting, grouping addresses by website and choosing thresholds without the test data. On held-out card data the Random Forest was the best model, with an area under the precision-recall curve of 0.800 (95% confidence interval 0.713 to 0.876), precision 0.926 and recall 0.789; oversampling before the split inflated the same model to 1.00. On web addresses a logistic model scored 0.999, but every legitimate address in the data was a secure site root starting with www and having no path; rewriting phishing addresses in the legitimate format cut its detection from 99.5% to 29.6%. A robustness gate now blocks such models, and a host-name model that replaced it detected 52% of phishing addresses at a 0.8% false-alarm rate. No labelled Shoprite fraud data exist yet, and the paper sets out how to collect them.

International Journal of Advanced Multidisciplinary Research and StudiesVol. 6(6)
Imbalanced Data Classification Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon