While it's non-trivial to inspect many aspects of tor traffic, an often used study metric has been the (determinable) percentage of connections to hidden services, which are usually assumed to be disproportionately malicious.
This ranges around ~5% across most studies, the most recent one I can find shows similar results[0].
Results are limited by the inability to account for lawful use of hidden services, but also the percentage of malicious use outside of them.