5
$\begingroup$

I'd like to produce 95% prediction intervals along with predictions from my model. I'm using a moderately large dataset and making thousands of predictions, so I was wondering if there was some way for me to calculate the prediction interval directly without bootstrapping.

I'm using the LinearRegression model (with multiple features) for now, though will most likely use one of the models that incorporate regularization down the road

$\endgroup$
3
  • $\begingroup$ A model based prediction interval is calculated using a (1+1/n)*residual-sd error bar for the predicted value $\endgroup$
    – AdamO
    Feb 8 '18 at 21:16
  • $\begingroup$ What model are you using? $\endgroup$
    – Danica
    Feb 8 '18 at 21:46
  • $\begingroup$ @Dougal I'm using the LinearRegression model (with multiple features) for now, though will most likely use one of the models that incorporate regularization down the road $\endgroup$ Feb 9 '18 at 15:12