Chapter 4 - Regression analysis

Abstract

Regression analysis is a method of choice whenever it predicts continuous variables given a multivariate input variable. It often provides explicit estimates of measure for the cause-effect relationship between the individual inputs and the outcome, with an error estimate given by an optimization algorithm. Regression models are used in various communication networks and the internet of things-related tasks. These models are mostly built using data-driven statistical machine learning techniques to explicitly obtain a parametric relationship between the input or independent variables and the output. The most common regression analysis class is the generalized linear models, including linear regression and logistic regression. Along with ridge regression and polynomial regression, we discuss the regression models in detail by establishing the theory and giving pseudocodes to view the programmer’s perspective while writing actual software for these simulation models. We address the problems inherent in the analysis of data and its impact on the choice of model, cost function, and optimization algorithm choice. We include a discussion of cross-validation for model selection as well as various regularization methods.