Whenever you're ready, walk me through how you'd approach this.
You are training a binary classifier where the positive class is much rarer than the negative class. A model with high accuracy may still miss most of the cases you care about.
How would you handle imbalanced datasets?