A proximity graph-based approach to the edited nearest neighbour rule
Sanchez JS, Pla F, Ferri FJ

In the context of Statistical Pattern Recognition, one of the most important subjects, with respect to training the sample set, consists of eliminating misclassified prototypes (Editing) using distance-based methods. In this paper, an attempt to use several proximity graphs -Gabriel Graph and Relative Neighbourhood Graph- for editing the Nearest Neighbour rule is presented. Experiments on synthetic and real data have been carried out in order to investigate the recognition accuracy of the edited reference set.. A comparison with other standard editing techniques (Wilson´s editing, Holdout editing and Multiedit) is also reported.