K Nearest Neighbour(kNN) Classifier #

• Works with numeric data

• Lazy learning

• Instance based, lazy evaluation model i.e. store the data, at times with minimal processing, and evaulted once the test tuple is received

Approaches #

• k-nearest neighbour approach - Instance represented as a point in euclidean space

• Case-based reasoning - Uses symbolic representations and knowledge-based inferences

• Target can be discrete or continuous

• Nearest neighbour is based on distance computations

• Consider all instances as members of n-dimensional space

Tuple Storage #

Brute Force

KD Trees - K Dimensional trees

Ball Trees - Binary Search Trees

• Classification cost is high in kNN since the testing phase is computationally intensive compared to the training phase.

Lazy: Less time in training and more in predicting