フリー問題

問題文

An insurer trains a model that flags claims for manual review. Only about one claim in two hundred is actually abusive, and the business states that missing an abusive claim costs far more than sending a clean claim to a reviewer by mistake. Which measure should the team watch while tuning?

選択肢

1. The squared residual that can punish the numeric misses, since the cost of each kind of mistake settles it.
2. The silhouette count that can compare the cluster widths, since the cost of each kind of mistake settles it.
3. The accuracy figure that can reward the majority predictions, since the cost of each kind of mistake settles it.
4. The recall indicator that can expose the missed detections, since the cost of each kind of mistake settles it.