Application of Multilevel Models in Classification and Regression Problems

There is a constant need to create methods for improving the quality indicators of information processing. In most practical cases, ranges target variables and predictors are formed under influence external internal factors. Phenomena such as concept drift cause model lose its completeness accuracy over time. The purpose work improve processing data samples based on multi-level models classification regression problems. A two-level architecture proposed. At lower level, analysis incoming flows sequences takes place, or tasks solved. upper divided into segments, current properties in subsamples determined, suitable lower-level assigned according achieved qualitative indicators. formal description given. order solving problems, sample preliminary carried out, model’s calculated, classifiers with best results determined. proposed solution makes it possible implement constantly learning systems. It aimed at reducing time spent retraining case transformation. Experimental studies were out several datasets. Numerical experiments have shown that can be considered an improvement ensemble flows. Training single classifier, rather than group complex models, reduce computational costs.