This document discusses using machine learning to select self-driving car test cases for simulation-based testing. It presents an approach called SDC-Scissor that uses ML models and road features extracted from test cases to predict if they will be safe or unsafe. This allows prioritizing potentially unsafe tests to improve testing efficiency. The approach is evaluated on its ability to identify safe and unsafe tests cases, and whether it can reduce testing time and costs. SDC-Scissor was able to select relevant test cases and showed potential to improve the cost-effectiveness of simulation-based self-driving car testing.