Monitoring Changes in Clustering Solutions: A Review of Models and Applications

Abstract

This article comprehensively reviews the applications and algorithms used for monitoring the evolution of clustering solutions in data streams. The clustering technique is an unsupervised learning problem that involves the identification of natural subgroups in a large dataset. In contrast to supervised learning models, clustering is a data mining technique that retrieves the hidden pattern in the input dataset. The clustering solution reflects the mechanism that leads to a high level of similarity between the items. A few applications include pattern recognition, knowledge discovery, and market segmentation. However, many modern-day applications generate streaming or temporal datasets over time, where the pattern is not stationary and may change over time. In the context of this article, change detection is the process of identifying differences in the cluster solutions obtained from streaming datasets at consecutive time points. In this paper, we briefly review the models/algorithms introduced in the literature to monitor clusters’ evolution in data streams. Monitoring the changes in clustering solutions in streaming datasets plays a vital role in policy-making and future prediction. Of course, it has a wide range of applications that cannot be covered in a single study, but some of the most common are highlighted in this article.