Automatic estimation the number of clusters in hierarchical data clustering

Abstract

Emergent pattern recognition is crucially needed for a real-time monitoring network to recognize emerging behavior of a physical system from sensor measurement data. To achieve effective emergent pattern recognition, one of the challenging problems is to determine the number of data clusters automatically. This paper studies the performance of the model-based clustering approach and using the knee of an evaluation graph for the estimation of the number of clusters. The working principle of these two methods is presented in the article. Both methods have been used for the classification of damage patterns for a benchmark civil structure. The performance of these two methods on determining the number of clusters and classification success rate is discussed.