'how to reverse rolling mean differencing after predicting values with ARIMA to get actual predicted value?
I am doing time series forecasting using ARIMA but the given data (36 months) is not stationary. To make it stationary I tried differencing, seasonal differencing and log transformation.
But only this worked:
roll_mean = df.Volume.rolling(window = 3).mean()
roll_mean_vol = df.Volume - roll_mean
adf, pval, usedlag, nobs, crit_vals, icbest = adfuller(roll_mean_vol.dropna().values)
print('ADF test statistic:', adf)
print('ADF p-values:', pval)
Output:-
ADF test statistic: -4.489204175344117
ADF p-values: 0.00020566352427521528
Now on predicting values for test dataset I get:
2021-09-30 -299.756094
2021-10-31 8.967911
2021-11-30 8.967911
2021-12-31 8.967911
2022-01-31 8.967911
2022-02-28 8.967911
2022-03-31 8.967911
Freq: M, dtype: float64
When my original data contains all positive values ranging from 2000 to 8000.
How do I reverse the effect of transformation to get actual predictions?
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|