Checking data informativity as the first step in data-driven modeling – case study


  • Amir Farzin
  • Kateryna Rabchuk
  • Bernt Lie
  • Nils-Olav Skeie



data-driven modeling, data informativity, Feature selection, system identification, industrial process, timestamp dataset


This paper reviews and introduces the strategies for testing a given dataset sampled from an unknown dynamic process to determine if it is sufficiently informative to model the system’s behavior. The presented tests should be done as the first step in data-driven modeling to avoid an endless search for a proper model which may not exist based on the available data. It is unrealistic that available data holds complete information about the system at hand. The tests also allow us to estimate how good the established model can be. Finally, the presented methodologies are applied to an actual process as the case study: modeling the decarbonization section in an ammonia plant.