Alternative EOI B - Interpretable HFT Crypto Anomaly Pipeline
Title (paste-ready): Interpretable Anomaly Detection for High-Frequency Cryptocurrency Markets: Open Tooling for Manipulation Forensics
EOI BODY (paste-ready)
High-frequency cryptocurrency markets process billions of trades daily across centralised and decentralised venues. Manipulation events propagate within seconds. Supervisors and exchange risk teams lack interpretable detection methods that explain each alert. Stephen Chan’s AUS Faculty Research Grant FRG23 produced labelled fraud cases through May 2025. Joerg Osterrieder published a reaction-times-to-macro-news methodology for high-frequency trading in 2025. Combining domain-labelled crypto data with a news-aligned attribution framework is rare and timely.
We propose evt-hft-anomaly, an open MIT-licensed library for interpretable detection of HFT crypto manipulation. The library streams on-chain order books, exchange tick data, and off-chain news arrivals. Extreme-value-theory anomaly detection runs on the streams with online Bayesian change-point gating. A diffusion-model layer drawn from AR-Pro (NeurIPS 2024) generates a counterfactual explanation for each alert. The methodological foundations include the team’s joint Physica A 2024 paper on metaverse NFT stylised facts plus the joint SSRN blockchain-anomaly primer with code. Foundation-model time-series tools support inference only. No model training is in scope.
Specific aims: (1) Extend FRG23 anomaly methods with regime-aware EVT and bivariate tail dependence. (2) Release a labelled HFT crypto manipulation event database, CC-BY year 1. (3) Deliver evt-hft-anomaly as a working library by month 18. (4) Validate on at least two crypto venues with measured detection latency under one minute. (5) Train one UAE-based PhD via AUS Office of Graduate Studies.
Application-readiness arc. Months 0-6 cover the Applied baseline with FRG23 codebase migration and dataset curation. Months 6-12 produce a working prototype on AUS HFT crypto data plus the first open-released event database. Months 12-18 reach Tech Development with cross-venue validation. Months 18-24 deliver the first Validation step. A named industry analyst from Joerg’s ING advisory channel or a COST 19130 European supervisor runs evt-hft-anomaly with measured latency by month 24.
PI Stephen Chan sits in AUS Mathematics & Statistics in Sharjah. Co-PI Joerg Osterrieder chairs COST Action 19130 with more than 300 researchers across 51 countries. Joerg also coordinates the EUR 3.8 million MSCA-DN Digital Finance Industrial Doctoral Network. Co-PI Youcheng Sun at MBZUAI Computer Science leads verification of the EVT detector and the diffusion-counterfactual layer, drawing on his Verifi paper (IEEE TDSC 2024). UAE delivery anchors at AUS plus MBZUAI through Co-PI presence. Two UAE-based PhDs are recruited, one at AUS under Stephen and one at MBZUAI under Youcheng. Capacity building includes the AUS course Interpretable AI for Tokenised Markets and workshops at AUS year 1 and MBZUAI year 2. All code MIT, methods/datasets CC-BY where licence allows.