A price model's evaluation should penalize occasional large numeric errors more strongly than MAE does. Which metric is a suitable candidate?

1. The number of model parameters alone
2. RMSE
3. Classification accuracy
4. The proportion of nonmissing input values