Understanding the Importance and Best Practices for EVENT_ID and ENTITY_ID in Amazon Fraud Detector

I'm working on implementing Amazon Fraud Detector and have some questions about the required EVENT_ID and ENTITY_ID.

1. What is the rationale behind requiring both EVENT_ID and ENTITY_ID?
2. Generating and storing these unique IDs can add complexity. Are there specific roles that they play in the fraud prediction model?
3. Are there any best practices for generating and managing these IDs?
4. What impact does marking ENTITY_ID as 'unknown' have on the prediction accuracy?

I've read the Fraud Predictions guide, but it lacks specific examples to clarify these points. Practical scenarios or sample solutions would be highly appreciated.

Thank you in advance for any insights!