A Comparative Study on E-commerce Fraud Detection Based on Ensemble Learning: Mechanism Diagnosis of SMOTE Failure in Imbalanced Low-dimensional Scenarios

Ziqi Jiang

doi:10.47297/wspciWSP2516-252745.20261002

PDF
Export
Share
Collection
Album

A Comparative Study on E-commerce Fraud Detection Based on Ensemble Learning: Mechanism Diagnosis of SMOTE Failure in Imbalanced Low-dimensional Scenarios

Volume 10, Issue 2

Views: 0 Downloads: 0

A Comparative Study on E-commerce Fraud Detection Based on Ensemble Learning: Mechanism Diagnosis of SMOTE Failure in Imbalanced Low-dimensional Scenarios
A Comparative Study on E-commerce Fraud Detection Based on Ensemble Learning: Mechanism Diagnosis of SMOTE Failure in Imbalanced Low-dimensional Scenarios
School of Computer Science and Technology, Tianjin University, Tianjin, 300354, China
Vol. 10, Issue 2, Pages: 213-220(2026)
Published：30 April 2026
DOI：10.47297/wspciWSP2516-252745.20261002
Accepted：

Scan QR Code

Jiang Ziqi.A Comparative Study on E-commerce Fraud Detection Based on Ensemble Learning: Mechanism Diagnosis of SMOTE Failure in Imbalanced Low-dimensional Scenarios[J].Creativity and Innovation,2026,10(02):213-220.
DOI：

Jiang Ziqi.A Comparative Study on E-commerce Fraud Detection Based on Ensemble Learning: Mechanism Diagnosis of SMOTE Failure in Imbalanced Low-dimensional Scenarios[J].Creativity and Innovation,2026,10(02):213-220. DOI： 10.47297/wspciWSP2516-252745.20261002.

摘要

Abstract

With the continuous expansion of the global e-commerce market

economic losses caused by transaction fraud have become increasingly severe

and fraud detection is essentially a highly imbalanced binary classification problem. This study systematically compares three mainstream ensemble learning algorithms (XGBoost

LightGBM

CatBoost) using a public Kaggle e-commerce fraud dataset (23

634 records

5.17% fraud rate). Nine comparative experiments were designed across three groups: baseline (default parameters)

tuned (hyperparameter optimization)

and SMOTE (oversampling). Experimental results show that hyperparameter tuning improved the average F1-score of the three models by 25.16%

with XGBoost achieving the best F1-score (0.3626) after tuning. However

SMOTE universally degraded the performance of all models

with the average F1-score dropping by 28.2% and Average Precision (AP) declining by approximately 29%. Through multi-dimensional diagnostic analysis

the root causes of SMOTE failure were identified as the extremely uneven distribution of fraud samples (coefficient of variation CV=1.1991)

the presence of a large number of categorical features

and SMOTE-induced feature importance distortion. Feature importance analysis further revealed that account age

transaction amount

transaction hour

and shipping address are stable and interpretable fraud indicators across models. This study provides clear model selection and sampling strategy recommendations for e-commerce fraud detection practice

and extends the application boundary of SMOTE from high-dimensional scenarios to low-dimensional but extremely unevenly distributed scenarios.

关键词

Keywords

references

Nilson Report. Global Card Fraud Losses[R]. Issue 1225, 2023.

Yang Y X. Research on Fraud Transaction Identification Based on Machine Learning[D]. China: University of Electronic Science and Technology of China, 2024.

Chen T, Guestrin C. XGBoost: A scalable tree boosting system[C]. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, USA, 2016: 785-794.

Hajek P, Olej V, Sokol P. Fraud Detection in Mobile Payment Systems using an XGBoost-based Framework[J]. Information Systems Frontiers, 2022, 24(6): 1889-1906.

Ke G, Meng Q, Finley T, et al. LightGBM: A highly efficient gradient boosting decision tree[C]. Advances in Neural Information Processing Systems, Long Beach, USA, 2017, 30: 3146-3154.

Chawla N V, Bowyer K W, Hall L O, et al. SMOTE: Synthetic minority over-sampling technique[J]. Journal of Artificial Intelligence Research, 2002, 16: 321-357.

Ramineni P, Mastouri R. Credit Card Fraud Detection Using CatBoost[C]. Proceedings of the 2022 International Conference on Machine Learning and Data Science, Bangalore, India, 2022: 123-130.

Fu Y, Dong B H. Research on E-commerce Spam Reviewer Identification[J]. Scientific Decision Making, 2015(9): 1-18.

Xiao Y, Li J, Zhang H. Application of Machine Learning Model in Fraud Identification: A Comparative Study of CatBoost, XGBoost and LightGBM[J]. Preprints.org, 2025, 2025(02): 1-19.

Blagus R, Lusa L. SMOTE for high-dimensional class-imbalanced data[J]. BMC Bioinformatics, 2013, 14(1): 106.

Zhang D Y. Research on User Payment Fraud Behavior Detection in E-commerce Transactions[D]. China: Beijing University of Posts and Telecommunications, 2023.

Yu S Y. Research on Hybrid Algorithm of Medical Insurance Fraud Detection Based on Random Forest[D]. China: Shandong University of Finance and Economics, 2023.

Kaggle. E-commerce Fraud Detection Dataset[EB/OL]. https://www.kaggle.com/, 2023.