Evaluation of a method for measuring the quality of SOM neural networks for the clustering task
Abstract
This work evaluates a method for selecting Self-Organizing Maps based on the quality of topological preservation and representation of the neural network input data for clustering benchmark data based on the segmentation of the neural network weights. We evaluated five clustering algorithms: k-means, hierarchical agglomerative, and three methods based on graph partitioning. The results showed that the method for selecting the best neural network was effective for all four databases evaluated, although it did not generate optimal results. We observed that the performance of the clustering algorithms varies according to the type of data, with k-means presenting good performance for hyperspherical data and for the Iris database, the agglomerative hierarchical method being more effective for the MNIST database, and a method based on graph partitioning being more effective for data with arbitrary structure.
References
Costa, J. A. and Netto, M. L. (2003). Segmentação do SOM baseada em particionamento de grafos. In VI Brazilian Conference on Neural Networks, page 451–456.
Delgado, S. et al. (2017). A SOM prototype-based cluster analysis methodology. Expert Systems with Applications, 88:14–28.
Deng, L. (2012). The mnist database of handwritten digit images for machine learning research. IEEE Signal Processing Magazine, 29(6):141–142.
Fisher, R. A. (1936). The use of multiple measurements in taxonomic problems. Annals of Eugenics, 7(2):179–188.
Hameed, A. A. et al. (2024). Improving the performance of self-organizing map using reweighted zero-attracting method. Alexandria Engineering Journal, 106:743–752.
Kaski, S. and Lagus, K. (1996). Comparing self-organizing maps. In International Conference on Artificial Neural Networks (ICANN), ICANN’96, Berlin, Heidelberg. Springer-Verlag.
Kohonen, T. (2001). Self-Organizing Maps. Springer Berlin, Heidelberg.
Melo Riveros, N. A., Cardenas Espitia, B. A., and Aparicio Pico, L. E. (2019). Comparison between k-means and Self-Organizing Maps algorithms used for diagnosis spinal column patients. Informatics in Medicine Unlocked, 16:100206.
Santos, L. et al. (2021). Quality control and class noise reduction of satellite image time series. ISPRS Journal of Photogrammetry and Remote Sensing, 177:75–88.
Silva, L. and Costa, J. A. (2011). A graph partitioning approach to SOM clustering. In Intelligent Data Engineering and Automated Learning - IDEAL 2011, pages 152–159.
Silva, M. A. S. d. et al. (2022). Tracking the connection between Brazilian agricultural diversity and native vegetation change by a machine learning approach. IEEE Lat. Am. T., 20(11):2371–2380.
Silva, M. A. S. d. et al. (2024). A Self-Organizing Map clustering approach to support territorial zoning. In Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications, pages 272–286. Springer Nature Switzerland.
Ultsch, A. et al. (1994). Knowledge extraction from artificial neural networks and applications. Parallele Datenverarbeitung mit dem Transputer, pages 148–162.
Vesanto, J. and Alhoniemi, E. (2000). Clustering of the Self-Organizing Map. IEEE Transactions on Neural Networks, 11(3):586–600.
Villmann, T., Der, R., Herrmann, M., and Martinetz, T. M. (1997). Topology preservation in self-organizing feature maps: Exact definition and measurement. IEEE TRANSACTIONS ON NEURAL NETWORKS, 8(2):256–266.
