¡@

Today, analyzing the user¡¦s behavior has gained wide importance in the data mining community. Typically, the behavior of a user is defined as a time series of his or her activities. In this paper, users are clustered based upon time series extracted from their behavior during the interaction with given system. Although there are several different techniques used to cluster time series and sequences, this paper will attack the problem by utilizing a novel incremental fuzzy clustering strategy in order to achieve the objective. Upon dimensionality reduction, time series data are pre-clustered using the longest common subsequence as an indicator for similarity measurement. Afterwards, by utilizing an efficient method, clusters are updated incrementally and periodically through a set of fuzzy approaches. In addition, we will present the benefits of the proposed system by implementing a real application: Customer Segmentation. In addition to having a low complexity, this approach can provide a deeper and more unique perspective for clustering of time series.

Keywords: data mining, incremental clustering, time series, fuzzy clustering, prototype

Retrieve PDF document (201207_03.pdf)

Received November 11, 2010; revised February 11, 2011; accepted April 16, 2011.
Communicated by Vincent S. Tseng.