Comparing and Detecting Stationarity and Dataset Shift

Abstract Machine learning algorithms have been increasingly applied to spatial numerical modeling. However, it is important understand when such methods will underperform. are impacted by dataset shift ; modeling domains of interest present non-stationarities there no guarantee that the trained models effective in unsampled areas. This work aims compare stationarity requirement geostatistical concept shift. Also, workflow developed detect data prior modeling, this involves applying a discriminative classifier and two sample Kolmogorv-Smirnov test model And, required lazy modification support vector regression proposed account for The benefits algorithm demonstrated on well-known non-stationary Walker Lake improves root mean squared error up 25% relative standard SVR approach, areas where present.