Fraud and Outlier Detection on Transaction Data
Finding rare suspicious records such as fraudulent payments or money laundering in tabular transaction data, with supervised or unsupervised detectors.
Tier Grades
Tier A: Frontier AI labs
feasible
Google, Mastercard and Featurespace deploy fraud and AML models at bank scale (HSBC AML AI).
Tier B: Indonesian AI labs
feasible
Unsupervised detectors such as ECOD need no labels or GPUs; access to bank data is not in the sources.
Tier C: Institution units such as ITB
feasible
ECOD came from a 6-author team with a parameter-free method.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Parameter-free detectors (ECOD) give a fast first proof of concept on client transactions; labels and data access decide the next step.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeECOD is parameter-free and estimates per-dimension empirical distributions, with no training compute to speak of.
- DataHSBC with Google Cloud AML AI: two to four times as much suspicious activity found as the previous system, with 60% fewer alerts.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 6 authors on ECOD.
- Time horizonYears.
- Shelf-life signalsBank-deployed AML models reporting fewer alerts (HSBC, 60%), Transaction embeddings from large behavioural models (Featurespace TallierLTM)
- Pipeline roleTier A and payment networks deploy production models; Tier C benchmarks detectors on Indonesian data; Tier D runs unsupervised detection proofs of value with banks and fintechs.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Carnegie Mellon University
Northeastern University
Featurespace
Google
HSBC
Mastercard
National University of Defense TechnologyLinked models
ECODtabularexemplarPyODtabularexemplarAnti Money Laundering AI (AML AI)tabularexemplarDecision Intelligence Protabularproduced byTallierLTMtabularproduced byDeep Isolation Forest (DIF)tabularproduced byCOPODtabularproduced byLOF (Local Outlier Factor)tabularproduced byOCSVM (One-Class SVM)tabularproduced byLinks
R&D TopicFraud and Outlier Detection on Transaction Data
Models 9
Anti Money Laundering AI (AML AI)Google, HSBC, tabularexemplarDecision Intelligence ProMastercard, tabularproduced byCOPODNortheastern University, Carnegie Mellon University, tabularproduced byECODNortheastern University, Carnegie Mellon University, tabularexemplarTallierLTMFeaturespace, tabularproduced byDeep Isolation Forest (DIF)National University of Defense Technology, tabularproduced byPyODCarnegie Mellon University, tabularexemplarLOF (Local Outlier Factor)tabularproduced byOCSVM (One-Class SVM)tabularproduced byPapers
- Deep Semi-Supervised Anomaly DetectionLukas Ruff et al., ICLR 2020 (per arXiv comment), 2019 founding
- Realistic Synthetic Financial Transactions for Anti-Money Laundering ModelsErik Altman et al., arXiv preprint, 2023 state of the art
- ADBench: Anomaly Detection BenchmarkSongqiao Han et al., NeurIPS 2022 (per arXiv comment), 2022 state of the art
- Enhancing Graph Neural Network-based Fraud Detectors against Camouflaged FraudstersYingtong Dou et al., CIKM 2020 (per arXiv comment), 2020 state of the art
- Deep Learning for Anomaly Detection: A SurveyRaghavendra Chalapathy et al., arXiv preprint, 2019 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2201.00382 https://arxiv.org/abs/2201.00382
- How HSBC fights money launderers with artificial intelligence | Google Cloud Blog cloud.google.com https://cloud.google.com/blog/topics/financial-services/how-hsbc-fights-money-launderers-with-artificial-intelligence
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Fraud and Outlier Detection on Transaction Data", https://aurion.semai.id/topics/fraud-and-outlier-detection-on-transaction-data/, records of October 2026.