Dynamic Selection of a Regression Model on the Base of Meta-Learning

In this paper, the main concepts of machine learning were investigated, the various types of machine learning algorithms were analyzed and also classification and regression models are considered. There is always a need for the improvement of model performance in machine learning. One of the ways to do so is the building of ensembles. Ensembles of models allow significantly improve the quality of model prediction. This work proposes the new method of the regression analyses - a dynamic method of choosing a regression model based on meta-learning. The model prediction is done by dynamically selecting a regression model for each new attribute from the dataset. The regression model for each attribute is selected using a classification model whose predictions are based on which models gave the least errors for each attribute in the training set. A new method of machine learning, as in the cases of ensembles, allows improving the predictions of the base models. The method has been tested on different data sets. A comparative analysis of a set of regression models with a new method was performed, which showed that the new method predicts better than the individual base models.