Catalogue of Artificial Intelligence Techniques
Aliases: Locally Weighted Linear Regression, k-Nearest Neighbour
Keywords: classification, instance, learning, nearest neighbour, patter, recognition
Categories: Pattern Recognition and Image Processing
Author(s): Alastair McFarlane
Instance-based learning is a term used to describe several methods , the simplest of which is mentioned below, used for the classification of data in patter recognition problems.
The way a new data point is classified depends on previous data which has been examined: the training set, for example.
The simplest instance-based approach is called the k-nearest neighbour algorithm. This algorithm plots data points in n-dimensional space (R n). The easiest example of this is 2-dimensional space, where all data points are plotted on the plane.
When a new data point is encountered, it is placed in the same class as it's “nearest neighbour”, measured as the actual distance between the two points in n-dimensional space (the “Euclidean Distance”)
- Mitchell, Tom M., Machine Learning, Ch 8. Instance-Based Learning, McGraw-Hill, 1997, pp.230--231.