Reporting performance measures for classification in percentage or fraction?

Question

I have seen classification metrics like f1-score, precision and recall being reported both as fractions and percentages. These measures are between 0 and 1 and represent ratios.

Can we report them as percentages or do we have to stick to fractions? For example:

1. When we have a precision of 0.5 can we say "The classifier has a precision of 50%." ?

2. Does it all account also for f1-score? Can we say "The classifier has a f1-score of 50%? F1-score is not exactly a ratio, but the harmonic mean of two ratios.

Answer 1

For anything that's truly a proportion, why not, as long as you are clear which of the two options you are using? The main reason to do one or the other is really just what is commonly done in a particular field.

Of course, it does not make sense to add a percentage sign for something that is in no way a proportion/percentage.

However, as you pointed out many of the performance measures that are between 0 and 1 do have a percentage interpretation (e.g. precision as the percentage of cases that the model classifies as "true" that are really in the "true"-class).

Sometimes the proportion interpretation is a bit obscure (=not commonly known/used) such as in the case of area under the receiver operator curve (percentage of all possible combinations of examples from the "true" and "false" class, for which the "true" example is ranked more highly than the "false" one). It's unusual to see AuROC reported as a percentage (so I would not normally do it), but it's not like it's wrong.