Here's the way I see it.
Place that person in a small sample size of the League player base. Let's say a million. Then shake it a bit. Now what are the criteria that you would search for, without knowing their name, that would not only catch this person, but catch other trolls and not accidentally catch people who are having an extremely bad run unintentionally?
Don't get me wrong, I think you should send a ticket in to Riot support about this person. But making a detection algorithm that needs to sift through millions of players without banning even 0.1% (which would be about 1000 people if we're using the small sample size number) incorrectly because they had really bad games, is not something that just happens.