How to calculate prediction intervals with LSTM's deterministic prediction?

How to calculate prediction interval/ predictive density with LSTM time series point forecast data?

Answer 1

[Aneela]

Hi Israt Fatema,

Monte Carlo Dropout technique can be used to estimate prediction intervals and predictive density for LSTM.

• Use Dropout in the LSTM layers not just during training but also during prediction for randomness.
• Make multiple predictions for the same input data with dropout enabled.
• The multiple distributions for each time step form an empirical distribution that can be used as Predictive Density.
• The variance in these predictions can be used to estimate the Prediction Interval.

Refer to the following MathWorks documentation for more information on Dropout layer:. https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.dropoutlayer.html