Machine Learning in MEV Detection

Project information

• Author: Jiayi Wang, Applied Mathematics, Class of 2024, Duke Kunshan University

• Instructor: Prof. Luyao Zhang, Duke Kunshan University

• Disclaimer: Submissions to the SRS2023 Innovate instructed by Prof. Luyao Zhang at Duke Kunshan University.

• Acknowledgments: I am deeply indebted to my professor Luyao Zhang for her invaluable patience and feedback. I could not started my journey without her instructions and help. Additionally, this endeavour would not have been possible without Flashbots. Lastly, I would be remiss in not mentioning my family, especially my parents.

• Project Summary: Miner Extractable Value (MEV) has garnered significant attention from researchers due to its implications for the security and privacy of blockchain networks. The current cryptocurrency market has difficulty solving and detecting the MEV in a timely manner, and the problem has become one of the biggest issues on Ethereum. According to Flashbots[1], MEV has extracted over $600 million in only two years, and the number is continuously increasing. Machine Learning has provided us with a method to identify the MEV in time. However, collecting a comprehensive dataset of labelled examples for different MEV strategies can be difficult, limiting the accuracy and generalizability of the models. Meanwhile, MEV strategies can evolve and adapt over time to avoid detection. Machine learning models need to continually update and adapt to new MEV techniques to remain effective. Adversarial strategies can also be employed to deliberately bypass detection methods, requiring constant vigilance and model refinement. The whole detection process requires several steps to complete. This project is meant to retrieve transaction data.

  • Methodology: web3 library

Table of Contents

Contents	URL
Code	https://github.com/SciEcon/SRS2023_MEV_Jay#Code
Data	https://github.com/SciEcon/SRS2023_MEV_Jay#Data
Data Source	Ethereum ETL
Spotlight	https://github.com/SciEcon/SRS2023_MEV_Jay#Spotlight
more about the author	https://github.com/SciEcon/SRS2023_MEV_Jay#more-about-the-author
references	https://github.com/SciEcon/SRS2023_MEV_Jay#references

Code

• export code

Data

• transaction data

Spotlight

Figure 1. Transaction Data Preview

Figure 1. Trace Data Preview

Figure 3. Traces over time

In Figure 3, generally, the count of traces is increasing. The rapid drop at the end is due to only half of the transactions in July 2023 are counted.

More about the Author

• headshot

Jiayi Wang is a junior student at Duke Kunshan University majoring in Applied Math & Computational Science tracking in Math. He has a solid foundation in Data Science and excels in Math. He has done research in trading strategies and portfolio construction in the Chinese Commodity Market. He is working under the guidance of Prof. Luyao Zhang on machine learning for Blockchain Security and Privacy.

References

• Piet, Julien, Jaiden Fairoze, and Nicholas Weaver. 2022. “Extracting Godl [Sic] from the Salt Mines: Ethereum Miners Extracting Value.” ArXiv:2203.15930 [Cs], March. https://arxiv.org/abs/2203.15930.
• Qin, Kaihua, Liyi Zhou, and Arthur Gervais. 2022. “Quantifying Blockchain Extractable Value: How Dark Is the Forest?” IEEE Xplore. May 1, 2022. https://doi.org/10.1109/SP46214.2022.9833734.

‌ ‌

Code Source

web3.py

Data Source

Ethereum ETL

Literature

• Piet, Julien, Jaiden Fairoze, and Nicholas Weaver. 2022. “Extracting Godl [Sic] from the Salt Mines: Ethereum Miners Extracting Value.” ArXiv:2203.15930 [Cs], March. https://arxiv.org/abs/2203.15930.
• Qin, Kaihua, Liyi Zhou, and Arthur Gervais. 2022. “Quantifying Blockchain Extractable Value: How Dark Is the Forest?” IEEE Xplore. May 1, 2022. https://doi.org/10.1109/SP46214.2022.9833734.