How would you evaluate a model for rare-event prediction when false negatives are especially costly for American Bureau Of Shipping?
