A fast approximately k-nearest-neighbour search algorithm for classification tasks

The k-nearest-neighbour (k-NN) search algorithm is widely used in pattern classification tasks. A large set of fast k-NN search algorithms have been developed in order to obtain lower error rates. Most of them are extensions of fast NN search algorithms where the condition of finding exactly the k nearest neighbours is imposed. All these algorithms calculate a number of distances that increases with k. Also, a vector-space representation is usually needed in these algorithms. If the condition of finding exactly the k nearest neighbours is relaxed, further reductions on the number of distance computations can be obtained. In this work we propose a modification of the LAESA (Linear Approximating and Eliminating Search Algorithm, a fast NN search algorithm for metric spaces) in order to use a certain neighbourhood for lowering error rates and reduce the number of distance computations at the same time.