Clustering Variable Length Sequences by Eigenvector Decomposition Using HMM

We present a novel clustering method using HMM parameter space and eigenvector decomposition. Unlike the existing methods, our algorithm can cluster both constant and variable length sequences without requiring normalization of data. We show that the number of clusters governs the number of eigenvectors used to span the feature similarity space. We are thus able to automatically compute the optimal number of clusters. We successfully show that the proposed method accurately clusters variable length sequences for various scenarios. 1 Motivation Although many algorithms exist for unsupervised classification of patterns into clusters, most of these methods require the data space X consists of ‘identical length’ data points (feature vectors) xi = (xi1, ..., xiN ) where N is the dimension of the data space, i.e. X : RN . Such algorithms include the ordinary implementations of decision trees, neural nets, Bayesian classifiers, ML-estimators, support vector machines, Gaussian mixture mo...

• Type: Conference
• Year: 2004
• Where: SSPR
• Authors: Fatih Murat Porikli
• Keywords: Clusters Variable Length, Data Space, Pattern Recognition, SSPR 2004, Variable Length Sequences