Statistical Learning from a Regression Perspective

There has been a good amount of development in the area of linear models and regression analysis. Advancements in computing facilities have led to the development of alternative techniques of modelling based on the philosophy of regression analysis. The emphasis of such approaches is more on computational aspects, concepts and algorithms than on developing only the hard core theory. Several such techniques have been developed and became popular in the last decade under the purview of topics of statistical learning. The present book compiles some of the recently developed statistical learning techniques. The emphasis in the book is more on the applications of such techniques through computations rather than on their mathematical and statistical properties. Nevertheless, the necessary mathematical and statistical details accompany the applications but they are not discussed in detail. This book should be of interest to statisticians for a few reasons. Firstly, it gathers together various techniques of statistical learning in one place and explains how to use them. Secondly, the author is well versed with the developments in the area of regression analysis; see Berk (2003). So he has successfully bridged the gaps between analytical and computational developments related to regression analysis. The book will serve as a base for such topics for a long time in spite of the rapid developments in the field of statistical learning. A key feature of this book is that the regression function is used in terms of conditional distributions. The book is developed in eight chapters. The discussion in Chapter 1 presents the need and motivation for learning the tools that are alternative to traditional theory of regression analysis through several data-based examples and with various convincing arguments. Some underlying concepts of regression analysis that are needed for the development of theory in further chapters are also explained with minimal mathematical input. Next, Chapter 2 elaborates various aspects of regression splines and regression smoothers. It presents stepwise discussion on different types of splines and penalized smoothing, and addresses issues that are related to various methods of choosing smoothers with different type of variables along with illustrations. Chapter 3 deals with the issues, concepts and methodology of classification and regression trees. The trio of statistical learning procedures—bagging, random forests and boosting—are discussed in Chapters 4, 5 and 6 respectively. Each of these chapters presents motivation, reasoning for using the methodology, steps involved and related issues in detail. The topic of …