Using Geometric Information in Prototype Selection for the Nearest Neighbour Rule
Sanchez JS, Pla F, Ferri FJ

An attempt to use the Gabriel and Relative Neighbourhood graphs for editing the Nearest Neighbour rule is presented, along with some extensions to apply editing and condensing in a combined way. Experiments on synthetic as well as real databases have been carried out in order to investigate the classification accuracy of the edited and edited-condensed sets using the approaches presented and other traditional methods. Results show that the schemes proposed outperform the classical editing algorithms, and are comparable to Hart's condensing.