Challenge #103: Just another game?

[cgoodman3]

Here's my solution:

Spoiler

Took a bit of time to understand the errors being generated by the regression tool, eventually realised it doesn't like null values.

Got quite a few offence metric in the prediction model, but guess that's important as that's how you score points. Being a Brit I don't know too much about NFL so hopefully haven't caused much disservice to the game!

[OllieClarke]

Spoiler

Took the 4 lowest correlated (non-expected) variables

[TonyA]

Solution attached. Looks like the neural network got me the closest. It hit both the 2016 results and 2017 results almost exactly but I also included the 2018 results and it didn't do quite so well. 

[KMiller]

Solution attached.

Spoiler

[arjanloogman]

Yes, I peeked...

For some reason, the Tree Model got me closest predicting the Super Bowl...?

Spoiler

 

Anyway, great to be learning this topic, slowly dawning on me...

...and I know it's close to blasphemous, but I even got the Pearson's correlation choose the variables for a predictive macro...

Spoiler

 

Kind regards,

Arjan Loogman

[hanykowska]

Nice predictive practice, although I'm not sure I did what was intended

Spoiler

[T_Willins]

Interesting to look at what data sets produce more accurate results.  In addition to predicting results using all data I broke out regular season from post season to see if it was more accurate (it was not), but not enough data for post season to return a result.  I found some feeds for NHL data, so I might try my hand at building predictive models for hockey.  I read an article that hockey is going to have about 3,000 data points per second for games, so lots to analyze.  

Spoiler

[JamesCameron]

Here's my effort - leaned heavily on what I learnt in challenge 18!

Spoiler