Neural Network Prediction Interval

Question

I created a neural network in Python for a regression problem. I would like to have a prediction intervals for each value. How would I go about approaching this since neural networks are nonlinear?

Answer 1

一种方法是计算验证集的残差，它将有一个分布，计算残差分布的均值、方差，如果您正在寻找 95% 将 +,- 2sigma 添加到您的预测中，这应该是您的预测区间。

Answer 2

My first though is using bootstrap and jackknife methodology. Using bootstrap method, you can compute empirical confidence interval for your prediction using sampling on your original dataset.

But you can find other answers in this question