"Riot took care of the easy problem first, so they can't possibly care about the hard problem."
If you pay attention you'd know there is feeder detection in the game, it's just far less mature than the language detection. Natural Language Processing is something that many different industries use, and adapting it to League is relatively simple. Compare that to trolling/intentional feeding, which is not only unique to video games but the behavior is different game to game. You couldn't take Troll Detection software for Minecraft and make it work on League, because trolls do different things in those games. Thus the entire system has to be built from scratch.
Because the entire system is built from scratch, it's more likely to be error prone. The training wheels are very much on this thing, as Riot doesn't want to ban someone falsely. As they get more confident in it, it will get to the same point as the language punishments.