Why are predicted points so arbitrary?
I have been wondering about this for a while. Predictions seem to be rather independent of actual player performance. I get that they can be wildly off at the beginning of the season, but they should at least a little bit look at the actual player performance.
Here is an example (best game format): Currently, I see "Arrow" with a predicted points of 23 for week 8. I have absolutely no idea how a human or algorithm (if it works) would come up with that prediction. If we look at some stats here: Arrow has an average of 53.14 for two games (2x 26.57). In direct comparison with these opponents he had 50.28 points (28.12 + 22.16). His lowest week this split was 36.56 (week 2 with 14.40 + 22.16). Just assume his worst performance will happen twice in one week he'd get 28.8 points (2x 14.40).
All of these values are far above the predicted 23 points.
Now I do get that you rather predict the underlying scores (such as kills, deaths, assists) and then calculate the score as usual, but the same issue happens with the individual stats. (I'm not gonna write those out here, a bit too much work as they don't show up for individual games in the stats table.)
So what generates these predictions? This is puzzling me. Any insight would be appreciated.