ISSN:
1433-755X
Keywords:
Cluster analysis
;
K-means
;
Monte Carlo
;
Moving method
;
Unsupervised learning
Source:
Springer Online Journal Archives 1860-2000
Topics:
Computer Science
Notes:
Abstract In recent years there has been increasing interest in the comparative clustering abilities of k-means, moving methods and self-organising neural networks. However, most comparative studies have either been restricted to specific problem areas or have been conducted with other limitations that do not provide a more general evaluation of the relative abilities of these methods under a wide variety of conditions. This report provides a systematic empirical evaluation of the clustering abilities of k-means, moving methods and two commonly used self-organising neural network architectures. Monte Carlo simulation examining the effects of cluster shape, dimensionality, noise, dispersion and number of clusters in the data is used to evaluate the above methods. Results indicate that, on average, k-means, moving methods and ‘winner take all’ self-organising networks perform equally well in terms of clustering ability. However, as the moving method consistently converges faster than k-means, under circumstances where convergence speed is an important factor it may well represent a more appropriate benchmark for future comparisons between pattern partitioning methods.
Type of Medium:
Electronic Resource
URL:
http://dx.doi.org/10.1007/BF01237937
Permalink