[FIXED] Is it necessary to discard outliers before applying LSTM on time series

Issue

I am trying to detect anomalies on a time series that controls battery voltage output. I find that my original dataset has some outliers. In this case do I need to remove those points using InterQuartile Range (IQR) or Zscore? of course before using the LSTM keras model

Solution

Removing or not removing outliers all depends on what you are trying to achieve. You write here that your goal is anomaly detection, so at first glance, it seems like a poor idea to remove points or values that you are trying to detect. However, if you detect values that are of such a nature that they cannot even be due to plausible anomalies, then yes, you should remove them. In all other cases, you should consider to keep them.

Answered By - Serge de Gosson de Varennes