Here are some reports on a few predictors:

  • Some are really good (device=hourlywindspeed+feeder+outagemonth with accuracy of 0.979)
  • Some are good (feeder=equipment+outagemonth@wheatbelt with acc of 0.773)
  • Some are not so good (outagemonth=feeder+outagecause@wheatbelt acc of 0.66)

Lower Acc often means either:

  • the data is noisy,
  • the designated features (x) and the designated response (y) are not relatable, and or
  • too few or too many features

In the given data set, no matter how hard the model tries…

All the predictors must go through a fair amount of analysis to make sure they are worthy of release
to production. What we did in the demo is a partial work.

Here is a sample of few…