- 42
- 2
A Phystech graduate has developed a machine learning algorithm that can identify and block sybil accounts created by scammers to steal cryptocurrency tokens given away as part of promotional campaigns.
The algorithm was tested on 2.5 million crypto wallets and demonstrated a 90% detection accuracy—twice as high as similar algorithms used in the crypto industry to protect airdrop campaigns from malicious attacks.
To illegally obtain rewards offered for promoting crypto projects, a scammer can create an entire network of fake wallets (sybil accounts). Such abuses distort metrics, trigger a decline in the token price, and ultimately undermine trust in the project.
[td]"My algorithm analyzes dozens of parameters: from behavioral patterns and cross-chain activity to network connections between wallets," explained Alexey Saplin, author of the thesis. "This allows us to identify even complex clusters that go undetected using standard methods. The algorithm demonstrated 90% accuracy, while most existing solutions demonstrate efficiency levels of 45–60%."[/td]The development was tested as part of an open competition organized by Layer Zero, thanks to which the project was able to reverse the unfair distribution of tokens worth $10.2 million.
Saplin's ML algorithm can be adapted for other crypto projects; work in this direction is already underway at MIPT. The author plans to continue his research in graduate school and hopes to ultimately create a universal tool for detecting fraudulent schemes in various blockchain ecosystems.
The algorithm was tested on 2.5 million crypto wallets and demonstrated a 90% detection accuracy—twice as high as similar algorithms used in the crypto industry to protect airdrop campaigns from malicious attacks.
To illegally obtain rewards offered for promoting crypto projects, a scammer can create an entire network of fake wallets (sybil accounts). Such abuses distort metrics, trigger a decline in the token price, and ultimately undermine trust in the project.
Saplin's ML algorithm can be adapted for other crypto projects; work in this direction is already underway at MIPT. The author plans to continue his research in graduate school and hopes to ultimately create a universal tool for detecting fraudulent schemes in various blockchain ecosystems.