Differential evolution-based weighted combination  of distance metrics for k-means clustering

Abstract

Bio-inspired optimization algorithms have been successfully used to solve many problems in engineering, science, and economics. In computer science bio-inspired optimization has different applications in different domains such as software engineering, networks, data mining, and many others. One of the main tasks in data mining is clustering, namely k-means clustering. Distance metrics are at the heart of all data mining tasks. In this paper we present a new method which applies differential evolution, one of the main bio-inspired optimization algorithms, on a time series k-means clustering task to set the weights of the distance metrics used in a combination that is used to cluster the time series. The weights are obtained by applying an optimization process that gives optimal clustering quality. We show through extensive experiments how this optimized combination outperforms all the other stand-alone distance metrics, all by keeping the same low complexity of the distance metrics used in the combination.