Aggregation of Individual Feature Based Similarities and Application to Hierarchical Clustering

• Published in: Chemical engineering transactions Vol. 46
• Main Authors: N. Zhou, B.J. Xie, T. Wang
• Format: Journal Article
• Language: English
• Published: AIDIC Servizi S.r.l 01-12-2015
• ISSN: 2283-9216
• DOI: 10.3303/CET1546037

The measure of similarity/dissimilarity is important to clustering algorithms. By using different similarity metrics, a clustering algorithm may achieve different clustering results. Many existed clustering methods employ classic distance measures such as the Euclidean and Manhattan distances as the measures of dissimilarity. In this paper, OWA aggregation operators with learned weighting vectors, such as DOWA and kNN-DOW A, are employed to aggregate the feature-based similarities between instances. The aggregated similarities provide more options in classic clustering algorithms and hence, increase their flexibility. The performances of proposed methods are tested in the classic hierarchical clustering. Experimental results shows that the DOWA and kNN-DOW A aggregated similarities have achieves better clustering accuracies than the Euclidean and Manhattan distances in hierarchical clustering.