Centroid stability with K-means fast learning artificial neural networks

This paper presents improvements made to the K-means fast learning artificial neural network (K-FLANN) to solve pattern classification problems. The latest improvements in K-FLANN, stabilizes the cluster formations such that the cluster centroids remain relatively consistent even though the data presentation sequence (DPS) changes. Previous implementations of FLANN experienced inconsistent cluster centroids that varied with DPS. The paper also discusses the selection criteria of parameter changes, outlining the important behavioral characteristics of the network as these parameters change. Experimental results show that the improved K-FLANN is resilient to changes in data presentation sequences (DPS) and preserves the clustering consistencies. It can also be used as a forced learning algorithm.