I have a time series dataset that does not have the stationary property. The dataset is monotonically increasing or sometimes showing no change over periods of time. Can I apply the ARIMA model to such datasets which do not have stationary properties? And if yes, what are the methods to process the data before feeding it into the model?
Topic arima time-series dataset
Category Data Science
If you are certain that the data is non-stationary, short answer is NO.
By applying ARIMA, I am assuming, you mean fitting ARIMA class of model to your data. Fitting the data is just a mathematical exercise and can be done on any dataset. The point is whether the inference can be made on such models?
So as such the inference can be drawn but the limiting distribution of estimated coefficient will not be standard (normal) unlike in stationary. As the t-values thrown by usual softwares are under the assumption of stationarity, this makes these ARIMA models inapplicable.
On how to process this data. That completely depends on type of non-stationarity. The most common answer you may find on Google is to difference the data but that is the right approach for unit root non-stationarity (and perhaps for linear time trends as well?). Your description of the data does not indicate towards unit-root.
Perhaps check for seasonality and deterministic trends. There are various other considerations also but for that a look at data is first required.
Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.