問題文
A regional insurer wants to shorten the time it takes to settle small motor claims. The volume is high, the rules that adjusters apply today are written down and stable, and the settlement amount must be explainable to the policyholder on request. An auditor is asked whether a machine learning model is the right choice for this requirement. Which consideration should carry the most weight in that advice?
選択肢
- Whether the vendor's benchmark accuracy exceeds the accuracy the adjusters achieve today.
- Whether the training data can be stored in the same data center as the claims system, because keeping the two together simplifies the network design.
- Whether the decision logic is already stable and written down, since a documented rule set can be automated without introducing a model whose behavior has to be re-evaluated after every retraining.
- Whether the model can be trained on a graphics processor, since training on general purpose hardware would take longer than the fixed delivery date allows.