Greetings fellow DeepLearners, How are we going to deal with outliers in our data and what kind of impact do they have. When viewing the Differenced data(the one with no trends and seasonality but noise) does removing the outliers here help us in the overall forecasting process?
Thank you for taking your precious time to read my question!
Have you seen this link?
Hi there,
in addition I can recommend to take a look at this article here:
In general, I would recommend to incorporate your domain knowledge in your strategy how to deal w/ outliers so that after your operations your data set is still representative of the problem that you want to solve. Visualization usually helps a lot! (E.g. it you know that the population of your feature that obtains some outliers is normally distributed that should hold true after your operation steps handling the outliers, like „e.g. sigma clipping“ https://i.stack.imgur.com/Cgmr4.png )
I would also suggest to understand the reason of outliers, e.g. with respect to if there are systematic reasons why outliers occur, e.g. due to limitations of the sensor, measurement equipment or something domain specific which can be improved in the future.
More into can be found here:
Best
Christian