Dataset segmentation considering the information about impact factors

Introduction: The application of machine learning methods involves the collection and processing data which comes from recording elements in offline mode. Most models are trained on historical then used forecasting, classification, search for influencing factors or impacts, state analysis. In long run, value ranges can change, affecting quality classification algorithms leading to situation when should be constantly readjusted taking into account input data. Purpose: Development a technique improve dynamically changing non-stationary environment where distribution change over time. Methods: Splitting (segmentation) multiple based information about target variables. Results: A segmentation has been proposed, affect ranges. Impact detection makes it possible form samples current alleged situations. Using PowerSupply dataset as an example, mass is split subsets considering effects external impacts formalized production rules. using membership function (indicator function) shown. sample divided finite number non-intersecting measurable subsets. Experimental values neural network loss shown proposed selected dataset. Qualitative indicators (Accuracy, AUC, F-measure) various classifiers presented. Practical relevance: results development methods. conditions functioning.