Fig. 5
From: NSPLformer: exploration of non-stationary progressively learning model for time series prediction
Relative stationarity defined as the ratio of the ADF test statistic between the model prediction and ground truth. From left to right, the dataset becomes progressively less stationary (i.e., the dataset becomes increasingly non-stationary). Models using only stationary methods (left) tend to output overly-stationarization time series, while our method (right) provides predictions that are much closer to the smoothness of the actual values.
