Approaches to Accuracy Improvement for Regression-Based Models

Abstract

The development of technologies is growing rapidly. Among these technologies, the Internet of Things and broadband data networks occupy a decisive place on the path to sustainable development. They allow monitoring, collecting, and transmitting large amounts of data. Today, data are the basis for decision-making. The development of computing systems allows implementing more complex data processing methods based on artificial intelligence. Regression models are widely used to solve forecasting problems in the field of machine learning. The diversity of data structures encourages the search for more accurate approaches and technologies. This paper is devoted to the development and analysis of three approaches to improving the accuracy of linear regression models. These approaches are based on individual fitting and use optimal segmentation, consideration of heteroscedasticity, and alternative regression analysis methods. The developed approaches can be effectively applied to improve the accuracy of forecasting in data analysis problems, in particular in economics, engineering, reliability and operation theory, and other areas of human activity.