Authors:
Huaying Li
and
Aleksandar Jeremic
Affiliation:
McMaster University Hamilton, Canada
Keyword(s):
Clustering, Information Fusion, Cluster Ensemble and Semi-supervised Learning.
Abstract:
In the recent years there has been tremendous development of data acquisition system resulting in a whole
new set of so called big data problems. Since these data structures are inherently dynamic and constantly
changing the number of clusters is usually unknown. Furthermore the ”true” number of clusters can depend
on the constraints and/or perception (biases) set by experts, users, customers, etc., which can also change.
In this paper we propose a new cluster detection algorithm based on a semi-supervised clustering ensemble
method. Information fusion techniques have been widely applied in many applications including clustering,
classification, detection, etc. Although clustering is unsupervised and it does not require any training data,
in many applications, expert opinions are usually available to label a portion of data observations. These
labels can be viewed as the guidance information to combine the cluster labels that are generated by different
local clusters. It consists of
two major steps: the base clustering generation and the fusion. Since the step of
generating base clusterings is unsupervised and the step of combining base clusterings is supervised, in the
context of this paper, we name the algorithm as the semi-supervised clustering ensemble algorithm. We then
propose to detect a new cluster utilizing the average association vector computed for each data point by the
semi-supervised method.
(More)