I am getting some encouraging results with the synaptic neural network software (
https://caza.la/synaptic/#/)
I have been using mvfc statistics from back in the years 2016 to 2019 to predict the outcomes of the games during those seasons.
So far, my training data has been minimal, just the team averages for the offense, defense, special teams, and other stat categories similar to the data in the Pickem StatRank tables here ->
http://ip-198-12-248-7.ip.secureserver. ... c/statRank
Here are the results from a sample run:
{
'2016': { wins: 33, losses: 7, marginDiff: 297, mDAve: '7.42' },
'2017': { wins: 33, losses: 7, marginDiff: 386, mDAve: '9.65' },
'2018': { wins: 30, losses: 10, marginDiff: 303, mDAve: '7.58' },
'2019': { wins: 35, losses: 5, marginDiff: 356, mDAve: '8.90' }
}
Obviously, some stats are more important than others in determing the outcome of a game. What this neural net software does is come up with the best weights to use for each stat in order to produce the desired result. The goal is to maximize the wins while minimizing the mdAve which is basically the difference between the predicted point spread and the actual game point spread. Results will hopefully get even better when I add in detailed stats for the categories but we shall see. These mdAve are quite a bit better than the AMDs (Average Margin Diffs) I am seeing from players in the Pickem game. This is incredibly powerful stuff and comes into play presently in things like self-driving vehicles, automated facial recognition, sales forecasting, marketing strategies, and many other things.
)