---
title: Detecting Order Flow Toxicity in Cryptocurrency Perpetual Futures: A LightGBM Approach Using L2 Microstructure Features
url: https://www.ml-quant.com/papers/ssrn/7704859/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-11-20
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: SSRN 7704859
source_url: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7704859
featured: 2026-11-20
citations: unknown
topic: Trading, Microstructure & Execution
---


# Detecting Order Flow Toxicity in Cryptocurrency Perpetual Futures: A LightGBM Approach Using L2 Microstructure Features

A machine-learning detector predicts toxic trades in cryptocurrency perpetual futures using L2 microstructure features, achieving 0.66 AUC-ROC and removing 12.9% of toxic trades.

- Source: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7704859
- Identifier: SSRN 7704859
- Released: 2026-11-18
- First featured: Quant Letter No. 140 (2026-11-20): https://www.ml-quant.com/issues/2026-11-20/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Trading, Microstructure & Execution
- Authors: Arkadii Skvorcov

## Related

- [SwingGuard: Risk-Conditioned Multi-Agent Swing Trading with Dual-Timescale Governance](https://www.ml-quant.com/papers/arxiv/2610.12988/): A dual-timescale multi-agent framework generates trading strategies conditioned on current risk and maintains ongoing position governance between strategy revisions on cryptocurrency, preserving returns after transaction costs.
- [Learning the Spoofability of Limit Order Books With Interpretable Probabilistic Neural Networks](https://www.ml-quant.com/papers/arxiv/2504.15908/): The article discusses a new real-time detection model for spotting potential market manipulation in cryptocurrency exchanges, with 31% of large orders identified as potential spoofs.
- [Deep reinforcement learning for arbitrage in decentralized exchanges](https://www.ml-quant.com/papers/ssrn/4666504/): The study explores trading performances under arbitrage conditions in decentralized exchanges, using a simulation model and deep reinforcement learning to determine optimal arbitrage strategies for eight cryptocurrency pairs.
- [Does a Trading Policy Need to Know What It Trades? Microstructure Invariance and Zero-Shot Transfer of Execution Policies](https://www.ml-quant.com/papers/arxiv/2611.08701/): An execution policy trained once on 954 stocks in normalized units generalizes to unseen assets, exchanges, and crypto with minimal performance loss.
- [Talking About Stocks: Pathways into Financial Engagement](https://www.ml-quant.com/papers/repec/ajk-ajkdps-431/): Tracking 6.8 million Reddit users reveals distinct entry routes to stock investing from budgeting, crypto, or gambling forums, with composition predicting market price and volume reactions.
- [A game-theoretic analysis of the market microstructure in decentralized exchanges with concentrated liquidity](https://www.ml-quant.com/papers/arxiv/2610.28863/): The research models liquidity provision on decentralized exchanges as a mean-field game and derives closed-form solutions linking critical volatility to market microstructure parameters, validated on Ethereum data.
