I have a pandas dataframe univariable time series data. The values of the column range between -1 and 1. However, it is left skewed with some outlier values greater than 0.5.
I want to do time series forecasting on this dataset but the predict result has been poor irrespective of the parameter tuning. I believe one issue might be with raw data - although it is between -1 and 1, the outliers might be making it difficult to predict (note these 'outliers' are valid occurrences of the event).
My question: is there a way to normalise the data such that the extreme values will not be pronounced? I tried log-normalization but wouldn't work with negative values and other scaling methods tend to keep the distribution.
question from:
https://stackoverflow.com/questions/65841490/normalize-pandas-column-with-skewed-values 与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…