Suppose you are building a classifier to detect harmful TikTok content. The positive class is very rare. Which metrics would you use to evaluate the model, and why?

"The overall interview process was fairly structured. It started with a recruiter screening to discuss my background, experience, motivation for the role, and general logistics. This was followed by technical interviews that covered a combination of coding, machine learning fundamentals, and discussion of previous projects. There were also questions related to applying machine learning to Trust & Safety problems. What went well was the discussion around my previous machine learning work and how I approached real-world modeling problems. I also found the interviewers professional and the questions generally relevant to the role. The more challenging part was the coding portion, especially solving algorithmic problems under time pressure while also clearly explaining the approach and complexity."