Should I remove the trend from timeseries when using DeepAR

I saw that for some other algorithms for timeseries data it is advised to remove trend and seasonality before doing the prediction (ex: ARIMA and LSTM)

I figured out from the paper that SageMaker's DeepAR deals internally with seasonality, but does the same thing stands for trend?
Let's say I have multiple timeseries, where some of them have positive, and some have negative trend. Should I remove trend and then use DeepAR prediction, or should I just ignore it and let DeepAR handle it?

Topic rnn preprocessing aws machine-learning

Category Data Science


Yes, you should.

The main reason you should do this is because when data is trending up/down, it's more difficult to sample the useful data when training the model as the data is changing constantly, lstm-like model would work if it's seen enough input/output patterns when training and those patterns reoccur when predicting. However, with trending data, the data near present(used when predicting) is quite different to those in the past(used when training).

That being said, not all time series forecasting models are subject to this issue, typically, NBeats can handle this well as it learns the parameter of the trend(may be linearity) which can be extended to the future without difficulties.

About

Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.