K-Nearest Neighbors

Abstract

This chapter gives an introduction to pattern recognition and machine learning via K-nearest neighbors. Nearest neighbor methods will have an important part to play in this book. The chapter starts with an introduction to foundations in machine learning and decision theory with a focus on classification and regression. For the model selection problem, basic methods like cross-validation are introduced. Nearest neighbor methods are based on the labels of the K-nearest patterns in data space. As local methods, nearest neighbor techniques are known to be strong in case of large data sets and low dimensions. Variants for multi-label classification, regression, and semi supervised learning settings allow the application to a broad spectrum of machine learning problems. Decision theory gives valuable insights into the characteristics of nearest neighbor learning results.