Quote:
Originally Posted by Malachii
Does your model incorporate any type of power rating for fighters? I know that it tends to significantly favor underdogs, but do you mind sharing a little bit more info about what else it looks for?
Yeah. Since the beginning an Elo component has always been a part of the model. The idea was that, using some sort of rating system, it would set the baseline. Then other factors would modify that. Seems like a natural way to think about things. So someone with a high rating and good style matchup would be more favored than their rating implied, etc..
I'm honestly not sure if that's a good or bad idea at this point. The problem I have I think is that the more layers I add the more squished towards 50/50 the predictions tend to become (i.e. the more underdogs I'm betting). I might need to really reexamine how I judge fighter styles, and what is causing this to happen.
Other than that it contains stuff you'd expect. Age, guessing the amount of wear and tear, if they are improving or declining, etc..