First of all, lets clarify that this is a hard problem to fully automate. This is because at its core, banning intentional feeding is basically giving a game as input and trying to tease out intent, whether someone was playing badly to grief the team or not. Also, the cost of a false positive is high, as it means banning innocent players who happened to play badly and alienating them from the game. Also, there are some people who try to skirt the system to try to avoid a ban while griefing their team. Keep in mind that there are tons of games, so looking at all of them manually is not feasable, and having the community help like the old tribunal comes with its own challenges.
I don't think your game is a good representative example of the challenge that riot faces. All it seems to establish is that K/D/A alone is not a good heuristic. But there is much more data that riot can use when crafting their heuristic. Some of this is exposed at the end of game stats screen, like damage to champions.
Another part of the issue is that reports can't be fully trusted, as some people don't have a clear distinction in their minds between inting and poor play, or think that poor play should be reported and banned, etc.
To skirt around the problem of banning the innocent, I suspect that riot flags some accounts for an immediate ban (if they are too obvious) and some for manual review if they are reported and meet a specific heuristic. This allows them to automate some of the process, while preventing false positives by looking manually at the less clear cut cases.
