If you're going to utilize statistics in a definition, you'd still need to define the population which you derive the statistics from.
Example:
If I define smart humans by the following:
Smart humans are those who are in the top 50% of a standardized intelligence test.
Then I'll need to provide a definition of what a human is.
It's not possible to define the population to fit the statistics, since it's possible to arbitrarily change the population to fit the bill. In fact, many individuals within the classification will be excluded even though included, which is a contradiction.
Example (close to yours):
Take a group A.
Assume group A have a 50% rate of being able to use fire.
Say that a group is "A-ish" if the group have a 50% rate of being able to use fire, or higher.
Take one person from the original population and assume that he/she cannot use fire.
That group consisting of that person is then, since he/she have a 0% rate, not "A-ish".
Which is a contradiction, since that person is a subset of group A which was assumed to be "A-ish".