Unsupervised Classifier for Dynamic Systems Based on ISODATA

Abstract

This paper describes a method for designing an algorithm to classifier a dynamic set. Nowadays, the classifiers are well known, this work shows a combination of different techniques in order to get a better classifier. The Bayes theory is the initial point to get a KNN classifier. In the set of study there is not possible to calculate the classes or their parameters, so an unsupervised classifier based on KNN classifier is used for this approach. Finally, in the last step, it is made a modification in the seed for applying in a dynamic set. It is shown that this approach reduces the process time in 25% and improves the accuracy in 90%.