Improving Neural-Network Classifiers Using Nearest Neighbour Partitioning

Improving classification of k nearest neighbours by mapping points to a partition space using a neural network optimized by Particle Swarm Optimization where points of the same class are closer to each other and points of different classes are spread further apart.

The objective of optimization for NNP is to find a neural network that maps points in the same class close together, while spreading points from different classes as far away from each other as possible. The target function includes a facility to account for class weights for imbalanced data. NNP generates flexible partitions and arbitrarily shaped decision boundaries easily and increases the chance of finding potential neural-network classifiers.