When predicting continuous numeric values

When the dataset is balanced across classes

When monitoring model performance over time

When prediction errors are measured as percentages