How do ml classifiers detect cryptocurrency fraud?

Could you elaborate on how machine learning classifiers specifically detect cryptocurrency fraud? I'm interested in understanding the techniques and algorithms involved. Are there any specific features or patterns that classifiers focus on? Do they analyze transaction data, user behavior, or both? What challenges do practitioners encounter in this area? Do you have any examples of successful fraud detection cases? And finally, how do these classifiers adapt and improve over time to catch evolving fraud schemes?