International Journal of Computational
Intelligence Research (IJCIR)
Volume 2, Number 1 (2006)
A weighted deterministic annealing algorithm for data clustering
Yang Xulei, Song Qing
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 329978
Medical Center, Vanderbilt University, VU Station B, #351631, Nashville, TN, 37235
The deterministic annealing (DA) approach to clustering and its extensions has demonstrated substantial performance improvement over standard supervised and unsupervised learning methods. However, its performance will be severely distorted if there exists noise (or outlines) in the given data set. This paper proposes a new robust clustering method -- weighted deterministic annealing (WDA) algorithm, which attempts to solve the noise sensitivity problem by reformulating the source distribution of conventional DA approach. The proposed method generates different weight for different pattern: bigger weight for good (non-noisy) pattern and smaller weight for noisy pattern, which effectively reduced the effect of noise on the final partitions. The superiority of the proposed method is supported by simulation results.
Deterministic Annealing, Robust Clustering, Supervised and Unsupervised Learning, Noise Sensitivity.