Editing Nearest Neighbor Method

Samples in the overlapping region can confuse the algorithms, so just delete them.

How do we know if the samples are in the confusing zone?

• Pre-classification to detect samples in confusing zone using nearest neighbor.
• Then remove misclassified examples.

To see improvement in performance:

$$P_1^E(e)=\frac{P_1(e)}{2[1-P_1(e)]}$$

Note that $P_1^E$ is the asymptotic error after editing and $P_1(e)$ is the error before editing.