Paper: arXiv 2312.13896

Authors: Hugo Thimonier, Fabrice Popineau, Arpad Rimmel, Bich-Liên Doan, Fabrice Daniel

Abstract

This study explores the application of anomaly detection (AD) methods in imbalanced learning tasks, focusing on fraud detection using real online credit card payment data. We assess the performance of several recent AD methods and compare their effectiveness against standard supervised learning methods. Offering evidence of distribution shift within our dataset, we analyze its impact on the tested models’ performances. Our findings reveal that LightGBM exhibits significantly superior performance across all evaluated metrics but suffers more from distribution shifts than AD methods. Furthermore, our investigation reveals that LightGBM also captures the majority of frauds detected by AD methods. This observation challenges the potential benefits of ensemble methods to combine supervised, and AD approaches to enhance performance. In summary, this research provides practical insights into the utility of these techniques in real-world scenarios, showing LightGBM’s superiority in fraud detection while highlighting challenges related to distribution shifts.

Complexity vs Empirical Score

  • Math Complexity: 5.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper presents a rigorous empirical evaluation on real-world credit card transaction data, analyzing distribution shifts and comparing multiple models with robust metrics. While it introduces advanced concepts in anomaly detection and imbalanced learning, its mathematics remains largely conceptual and algorithmic rather than dense theoretical derivations.

Research Flowchart

  flowchart TD
  A["Research Goal: Compare AD Methods vs Supervised Learning<br>for Fraud Detection"] --> B["Dataset: Real Online Credit Card Data<br>Imbalanced & Distribution Shift Exists"]
  B --> C["Methodology: Evaluate Anomaly Detection<br>vs Standard Supervised Learning"]
  C --> D{"Computational Process"}
  D --> E["LightGBM Supervised Model"]
  D --> F["Anomaly Detection Methods"]
  E & F --> G["Key Outcomes"]
  G --> H["LightGBM Superior Performance"]
  G --> I["AD Methods More Robust<br>to Distribution Shift"]
  G --> J["LightGBM Captures<br>Most AD-Detected Frauds"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper empirically evaluates recent anomaly detection methods against LightGBM for fraud detection on real online credit card payment data.

Source sentence
Our findings reveal that LightGBM exhibits significantly superior performance across all evaluated metrics but suffers more from distribution shifts than AD methods.

Assumes 2

  • Only fraudulent behaviors change over time, so AD models should be more robust to distribution shift than supervised approaches. model
    Source sentence
    assuming only fraudulent behaviors change over time, AD models should be more robust to distribution shift than standard supervised approaches.
  • If the normal distribution is well characterized, AD models can always exclude new types of anomalies. model
    Source sentence
    if the normal distribution is well characterized, they should always be able to exclude new types of anomalies.

Data

  • Online credit card payments dataset · major French bank other proprietary
    2018-01 to 2021-12 · online credit card transactions
    Source sentence
    We dispose of a labeled dataset of online credit card payments made available by a major French bank. Our dataset contains 145 features

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • LightGBM consistently outperforms all tested AD methods across various evaluation metrics. performance
    Source sentence
    Our findings reveal that LightGBM exhibits significantly superior performance across all evaluated metrics but suffers more from distribution shifts than AD methods.
  • LightGBM suffers more performance degradation due to distribution shift than AD methods. negative
    Source sentence
    show that GBDT suffer significantly more from distribution shift than AD methods while displaying substantially better fraud detection performance than all tested AD methods.
  • LightGBM captures the majority of frauds detected by AD methods, limiting the benefit of ensembling. performance
    Source sentence
    LightGBM also captures the majority of frauds detected by AD methods. This observation challenges the potential benefits of ensemble methods to combine supervised, and AD approaches to enhance performance.
  • ECOD detected 20.41% of fraud cases not flagged by LightGBM, but 89.02% of ECOD‑detected frauds were also detected by LightGBM. performance
    Source sentence
    that ECOD did not flag. Notably, the 3.28% represents 10.98% of the fraud cases detected by ECOD. In other words, 89.02% of the fraud cases detected by ECOD were also detected by LightGBM.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 33 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (159)

  1. Detecting Semantic Anomalies (2019) — link
  2. Statistical Analysis of Nearest Neighbor Methods for Anomaly Detection (2019) — link
  3. Beyond Individual Input for Deep Anomaly Detection on Tabular Data (2023) — link
  4. Trac I n AD : Measuring Influence for Anomaly Detection (2022) — link
  5. Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion (2010) — link
  6. Contractive auto-encoders: Explicit invariance during feature extraction (2011)
  7. Contrastive Multiview Coding (2019) — link
  8. Momentum Contrast for Unsupervised Visual Representation Learning (2020)
and 151 more
  1. A Simple Framework for Contrastive Learning of Visual Representations (2020) — link
  2. PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentation (2021)
  3. The Unreasonable Effectiveness of Deep Features as a Perceptual Metric (2018)
  4. Memorizing Normality to Detect Anomaly: Memory-Augmented Deep Autoencoder for Unsupervised Anomaly Detection (2019)
  5. Fixing Bias in Reconstruction-based Anomaly Detection with Lipschitz Discriminators (2022) — link
  6. Graph Based Anomaly Detection and Description: A Survey (2015) — link
  7. Informal identification of outliers in medical data (2000)
  8. A research study on unsupervised machine learning algorithms for early fault detection in predictive maintenance (2018) — link
  9. A Probabilistic Resource Allocating Network for Novelty Detection (1994) — link
  10. On Estimation of a Probability Density Function and Mode (1962) — link
  11. Support Vector Method for Novelty Detection (1999)
  12. Support Vector Data Description (2004) — link
  13. Rule Extraction in Unsupervised Anomaly Detection for Model Explainability : Application to OneClass SVM (2020) — link
  14. TabR: Unlocking the Power of Retrieval-Augmented Tabular Deep Learning (2023) — link
  15. Classification-Based Anomaly Detection for General Data (2020) — link
  16. Why do tree-based models still outperform deep learning on typical tabular data? (2022) — link
  17. Revisiting Deep Learning Models for Tabular Data (2021) — link
  18. Self-Attention Between Datapoints: Going Beyond Individual Input-Output Pairs in Deep Learning (2021) — link
  19. XGBoost: A Scalable Tree Boosting System (2016) — link
  20. Well-tuned Simple Nets Excel on Tabular Datasets (2021)
  21. Certifiably Adversarially Robust Detection of Out-of-Distribution Data
  22. Certifiably Adversarially Robust Detection of Out-of-Distribution Data (2021) — link
  23. Bootstrap your own latent: A new approach to self-supervised Learning (2020) — link
  24. Exploring Simple Siamese Representation Learning (2020) — link
  25. Understanding self-supervised Learning Dynamics without Contrastive Pairs (2021) — link
  26. Using Self-Supervised Learning Can Improve Model Robustness and Uncertainty (2019)
  27. Deep Learning for Anomaly Detection : A Survey (2019) — link
  28. The Trace Criterion for Kernel Bandwidth Selection for Support Vector Data Description (2020) — link
  29. Unsupervised Anomaly Detection with Multi-scale Interpolated Gaussian Descriptors (2021) — link
  30. Simple and Effective Prevention of Mode Collapse in Deep One-Class Classification (2020) — link
  31. Anomaly Detection for Tabular Data with Internal Contrastive Learning (2022) — link
  32. Contrastive Predictive Coding for Anomaly Detection (2021) — link
  33. Deep Anomaly Detection on Attributed Networks (2019) — link
  34. Deep Multi-sphere Support Vector Data Description (2020) — link
  35. Towards Domain-Agnostic Contrastive Learning (2021) — link
  36. COPOD: Copula-Based Outlier Detection (2020) — link
  37. Fonctions de repartition an dimensions et leurs marges (1959)
  38. \ SSD \ : A Unified Framework for Self-Supervised Outlier Detection (2021) — link
  39. Domain-Agnostic Clustering with Self-Distillation (2021) — link
  40. DROCC: Deep Robust One-Class Classification (2020) — link
  41. Statistical Analysis of Nearest Neighbor Methods for Anomaly Detection
  42. Extended Isolation Forest (2021) — link
  43. A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks (2018) — link
  44. Deep Anomaly Detection with Outlier Exposure (2019) — link
  45. ESAD : End-to-end Deep Semi-supervised Anomaly Detection (2021) — link
  46. Tabular Data: Deep Learning is Not All You Need (2021) — link
  47. Adam: A Method for Stochastic Optimization (2015) — link
  48. Optuna: A Next-generation Hyperparameter Optimization Framework (2019)
  49. Scikit-learn: Machine Learning in P ython (2011)
  50. Isolation Forest (2008) — link
  51. RaPP: Novelty Detection with Reconstruction along Projection Pathway (2020) — link
  52. Backpropagated Gradient Representations for Anomaly Detection (2020) — link
  53. A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks (2018) — link
  54. SpecAE : Spectral AutoEncoder for Anomaly Detection in Attributed Networks (2019) — link
  55. Isolation Forest (2008) — link
  56. Open Category Detection with PAC Guarantees (2018) — link
  57. Explainable Deep One-Class Classification (2021) — link
  58. Hybrid Models with Deep and Invertible Features (2019) — link
  59. Deep Weakly-supervised Anomaly Detection (2020) — link
  60. SAINT: Improved Neural Networks for Tabular Data via Row Attention and Contrastive Pre-Training (2021) — link
  61. Lightgbm: A highly efficient gradient boosting decision tree (2017)
  62. The Relationship between Precision-Recall and ROC Curves (2006) — link
  63. One- Class Classification : A Survey (2021) — link
  64. Credit Card Fraud Detection in E- Commerce : An Outlier Detection Approach (2019) — link
  65. FACE : Feasible and Actionable Counterfactual Explanations (2020) — link
  66. Neural transformation learning for deep anomaly detection beyond images (2021)
  67. Deep One-Class Classification (2018) — link
  68. Deep Semi-Supervised Anomaly Detection (2020) — link
  69. Rethinking Assumptions in Deep Anomaly Detection (2020) — link
  70. Self- Attentive , Multi-Context One-Class Classification for Unsupervised Anomaly Detection on Text (2019) — link
  71. Self- Attentive , Multi-Context One-Class Classification for Unsupervised Anomaly Detection on Text (2019) — link
  72. A Unifying Review of Deep and Shallow Anomaly Detection (2021) — link
  73. Detecting Out-of-Distribution Examples with Gram Matrices
  74. Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery (2017) — link
  75. Detection of Anomalies in Large Scale Accounting Data Using Deep Autoencoder Networks (2018) — link
  76. Distributed Top-K Outlier Detection from Astronomy Catalogs using the DEMAC System (2007) — link
  77. In-Network PCA and Anomaly Detection (2006) — link
  78. Flow-Based SVDD for Anomaly Detection (2021) — link
  79. EdgeCentric : Anomaly Detection in Edge-Attributed Networks (2016) — link
  80. Efficient Anomaly Detection via Matrix Sketching (2018) — link
  81. Learning and Evaluating Representations for Deep One-class Classification (2021) — link
  82. Ellipsoidal Subspace Support Vector Data Description (2020) — link
  83. CSI : Novelty Detection via Contrastive Learning on Distributionally Shifted Instances (2020) — link
  84. Deep One-Class Classification (2018)
  85. APATE : A Novel Approach for Automated Credit Card Transaction Fraud Detection Using Network-Based Extensions (2015) — link
  86. Contrastive Training for Improved Out-of-Distribution Detection (2020) — link
  87. NetWalk : A Flexible Deep Embedding Approach for Anomaly Detection in Dynamic Networks (2018) — link
  88. Adversarial Self-Supervised Learning for Out-of-Domain Detection (2021) — link
  89. Deep Structured Energy Based Models for Anomaly Detection (2016)
  90. A Practical Guide to Training Restricted Boltzmann Machines (2012) — link
  91. Hybrid Models for Open Set Recognition (2020) — link
  92. Deep Residual Flow for Out of Distribution Detection (2020) — link
  93. DEEP AUTOENCODING GAUSSIAN MIXTURE MODEL FOR UNSUPERVISED ANOMALY DETECTION (2018)
  94. The Effect of Class Distribution on Classifier Learning: An Empirical Study (2001)
  95. A Multiple Resampling Method for Learning from Imbalanced Data Sets (2004) — link
  96. Editorial: Special Issue on Learning from Imbalanced Data Sets (2004) — link
  97. Experiments with a New Boosting Algorithm (1996)
  98. A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting (1997) — link
  99. Cost-Sensitive Boosting for Classification of Imbalanced Data (2007) — link
  100. Data Mining for Imbalanced Datasets: An Overview (2005) — link
  101. Issues in Mining Imbalanced Data Sets - A Review Paper (2005)
  102. Robust random cut forest based anomaly detection on streams (2016) — link
  103. PIDForest: Anomaly Detection and Certification via Partial Identification (2019) — link
  104. The Class Imbalance Problem: A Systematic Study (2002)
  105. Addressing the Curse of Imbalanced Training Sets: One-Sided Selection (2000)
  106. KNN Approach to Unbalanced Data Distributions: A Case Study Involving Information Extraction (2003)
  107. A Study of the Behavior of Several Methods for Balancing Machine Learning Training Data (2004) — link
  108. Applying One-Sided Selection to Unbalanced Datasets. (2000)
  109. On predicting rare classes with SVM ensembles in scene classification (2003) — link
  110. Learning and Evaluating Representations for Deep One-Class Classification (2021) — link
  111. CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances (2020)
  112. Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual (2020) — link
  113. Mean-Shifted Contrastive Loss for Anomaly Detection (2021) — link
  114. Towards Domain-Agnostic Contrastive Learning (2021) — link
  115. Class-boundary alignment for imbalanced dataset learning (2003)
  116. Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hypersphere (2020)
  117. Auto-Encoding Variational Bayes (2014) — link
  118. Nonlinear principal component analysis using autoassociative neural networks (1991)
  119. Identification of Outliers (1980)
  120. Memorizing Normality to Detect Anomaly: Memory-Augmented Deep Autoencoder for Unsupervised Anomaly Detection (2019)
  121. A robust anomaly finder based on autoencoders (2020) — link
  122. Neural Networks and Principal Component Analysis: Learning from Examples without Local Minima (1989) — link
  123. Autoencoders for unsupervised anomaly detection in high energy physics (2021) — link
  124. Addressing the Curse of Imbalanced Training Sets: One-Sided Selection (1997)
  125. UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction (2020) — link
  126. PyOD: A Python Toolbox for Scalable Outlier Detection (2019) — link
  127. Evaluating categorical encoding methods on a real credit card fraud detection database (2021) — link
  128. Visualizing Data using t-SNE (2008) — link
  129. SMOTE–IPF: Addressing the noisy and borderline examples problem in imbalanced classification by a re-sampling method with filtering (2015) — link
  130. Geometric Dataset Distances via Optimal Transport (2020) — link
  131. Learning from imbalanced data: open challenges and future directions (2016)
  132. Classification of imbalanced data: a review (2011) — link
  133. Neural Network Method for failure detection with skewed class distribution (2004) — link
  134. Applying Support Vector Machines to Imbalanced Data Sets (2004) — link
  135. Efficient algorithms for mining outliers from large data sets (2000)
  136. Fast outlier detection in high dimensional spaces (2002)
  137. Compounds Activity Prediction in Large Imbalanced Datasets with Substructural Relations Fingerprint and EEM (2015)
  138. Effective detection of sophisticated online banking fraud on extremely imbalanced data (2013) — link
  139. Generative Adversarial Nets (2014)
  140. SMOTE: Synthetic Minority Over-sampling Technique (2002) — link
  141. Learning from Imbalanced Data (2009) — link
  142. Exploratory Undersampling for Class-Imbalance Learning (2009) — link
  143. Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning (2005)
  144. ADASYN: Adaptive synthetic sampling approach for imbalanced learning (2008) — link
  145. LOF: Identifying Density-Based Local Outliers (2000) — link
  146. ECOD: Unsupervised Outlier Detection Using Empirical Cumulative Distribution Functions (2022) — link
  147. ECOD: Unsupervised Outlier Detection Using Empirical Cumulative Distribution Functions (2022) — link
  148. Extreme Re-Balancing for SVMs: A Case Study (2004) — link
  149. An Empirical Study on Class Rarity in Big Data (2018) — link
  150. Extended Isolation Forest (2021) — link
  151. Learning Long Term Style Preserving Blind Video Temporal Consistency (2021) — link

0 of 159 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.