Mastering pandas - Second Edition

KNN

K-nearest neighbors (KNN) is an unsupervised locality-based regression and classification method. It considers each row of data as a point in n-dimensional space and finds k similar (neighboring) points based on their distance (for example, Euclidean for numeric data and Hamming for categorical data). To find the value for that row and that column of data, it takes an average of all its neighboring rows for that column and assigns the average as a value.

To summarize, it can be said that it defines a locality around a point and calculates a local average instead of a global average. This makes sense most of the time and is used instead of the global average because neighborhood behavior is a better approximator of the behavior of a sample ...