The bank realizes that the model it selects to predict who will default on a loan is likely to be less accurate at predicting what the eventual defaulters will do than it will be at predicting what the people who eventually repay will do$.$ This is concerning because in practice it is more important to predict the defaults accurately than it is to predict the nondefaults accurately.$$ Which of the following are useful ways to reduce this problem?$$ Select all that apply

The bank realizes that the model it selects to predict who will default on a loan is likely to be less accurate at predicting what the eventual defaulters will do than it will be at predicting what the people who eventually repay will do$.$ This is concerning because in practice it is more important to predict the defaults accurately than it is to predict the nondefaults accurately.$$ Which of the following are useful ways to reduce this problem?$$ Select all that apply

Look at your records to see which type of algorithm has usually been the most accurate at predicting rare events.$$ Use the model you made with this algorithm regardless of what the percentages correct turn out to be for all the different models.

To create the models to predict default, use a training data set consisting only of actual defaulters so the models focus more closely on the specific characteristics of defaulters.

When using the final model to make predictions, consider an estimated probability of default of something less than $0\mathrm{.}5($say$,$ as low as $0\mathrm{.}3)$ to be a prediction of default.

Select the model to use based partly on its percentage correct for the defaulters, not solely on its overall percentage correct.