Metrics to determine model goodness.

[A143865]

How do we determine how good our model is?

• average difference of predicted number of yards vs actual number of yards?

As a benchmark can we use another model, like ESPN, and see how well that model predicts.

[sportygavin]

Yeah, my plan was to compare the percent error or difference of some sort between predicted and actual yards. I was then going to compare those from our model to ESPN's fantasy football model on a graph.

• Our model vs ESPN's model (either predicted yards or percent error)
• Predicted yards vs actual yards

And then, look closer at the games where the model was more accurate and other games where it wasn't, and see if there is a reason for that within the games. Possible see if ESPN has different metrics that make their model better.

[A143865]

Have you coded up this metric calculation yet? I think you might have the start of it; would like to check if I missed something.

[sportygavin]

No, I haven't yet. All I have is the mean squared error and mean absolute error. I still have to find the percent error for each of the games and then do that for the ESPN model as well.