Jean-Pierre Norguet's Review

Jean-Pierre Norguet's review of article "Data mining using rule extraction from Kohonen self-organising maps"

Paper reviewed: Malone J., McGarry K., Wermter S., Bowerman C., Data mining using rule extraction from Kohonen self-organising maps, Journal of Neural Computing and Applications, 15(1): 9-17, 2006. Review date: 22 June 2006. Review published with ACM Computing Reviews [http://www.reviews.com].

Review

This paper presents a novel approach to automatically recognizing the boundaries of complex n-dimensional data clusters. This approach requires a trained Kohonen self-organizing feature map, a particular kind of neural network. This kind of neural network produces a two-dimensional representation that supports supervised visual interpretation of the cluster boundaries. To support unsupervised interpretation of the cluster representation, this paper proposes a rule-extraction algorithm based on unified distance matrices, and data mining’s interestingness measures. This technique could be applied in enterprise systems, as well as in bioinformatics and medical systems. For example, the automated detection of cancers in medical images from radiographics and scanners could benefit from this technique.

The use of the LungCancer data set is assessed in the paper. Nevertheless, a strong understanding of neural networks, data mining, and knowledge discovery is needed to apply the technique. The use of the results is not covered; the rule sets provided by the technique can be provided to decision systems or human users. If the rule set formats are clearly suitable for decision systems, the intuitiveness of the results for human users remains unclear. Applying this technique with real users should, however, clarify this issue.

I found this paper to be an interesting articulation of complex techniques, and I believe that applying this technique to real-world situations would further reveal the concrete applicability of the approach.

Back to Jean-Pierre Norguet's homepage.