Methodological aspects of neural network clustering of regions: analysis of variability in self-organizing maps (Kohonen maps) results
Abstract
The purpose of the article is to identify and analyze the causes of variability in clustering results of Ukrainian regions by demographic indicators when applying the Self-Organizing Maps (Kohonen maps) method, as well as to develop methodological recommendations for interpreting and using neural network analysis results in human-geographical research.
Main material. The study is based on the analysis of demographic indicators of Ukrainian regions using the Self-Organizing Maps (SOM) method. Three clustering variants of Ukrainian regions were performed to identify stable and unstable cluster groups. Demographic indicators were normalized using linear scaling method with direct and inverse indexation formulas. Stable cluster cores were identified: Western Ukrainian regions with the best demographic situation (first group – Volyn, Rivne, Zakarpattya regions; second group – Ivano-Frankivsk, Lviv, Ternopyl, Chernivtsi regions); central-eastern industrial and industrial-agricultural regions with the worst demographic situation (Dnipropetrovsk, Zaporizhzhya, Mykolayiv, Sumy, Poltava, Kharkiv regions); predominantly central agricultural-industrial regions with unfavorable demographic situation (first group – Cherkasy and Chernihiv regions; second group – Vinnytsya and Khmelnytskiy regions). Regions with unstable cluster affiliation were determined (city Kyiv, Odesa, Kirovohrad, Zhytomyr, Kyiv, Kherson regions). Four main factors of SOM variability were analyzed: random initialization of neuron weights, stochastic nature of learning, nonlinearity and complexity of topological transformation, and the influence of network configuration parameters.
Conclusions. The variability of Self-Organizing Maps results is due to their stochastic nature, which is not a disadvantage but allows for the identification of multiple valid structures in data. To improve the reliability of results, it is recommended to perform multiple network training with cluster stability assessment and combine SOM with other territorial classification methods.
Downloads
References
Derzhavna sluzhba statystyky Ukrainy [State Statistics Service of Ukraine]. https://www.ukrstat.gov.ua/ [in Ukrainian].
Niemets, L. M., Barkova, H. A., & Niemets, К. А. (2009). Medychna haluz Kharkivskoi oblasti: terytorialni osoblyvosti, problemy ta shliakhy vdoskonalennia (suspilno-heohrafichni aspekty) [The medical sector of Kharkiv region: territorial features, problems and ways of improvement (human-geographical aspects). Kyiv, Ukraine : Chetverta khvylia [in Ukrainian].
Niemets, L. M., & Sehida, K. Yu. (Eds.) (2017). Innovatsiino-investytsiinyi potentsial yak osnova konkurentospromozhnosti rehionu (na prykladi Kharkivskoi oblasti) [The innovative-investment potential as the regional competitiveness base (a case study of Kharkiv region)]. Kharkiv, Ukraine : KhNU imeni V. N. Karazina [in Ukrainian].
Pylypenko, I. O., & Malchykova, D. S. (2007). Metodyky suspilno-heohrafichnykh doslidzhen (na materialakh Khersonskoi oblasti) [Methods of human-geographical research (based on the materials of Kherson region)]. Kherson : PP Vyshemyrskyi V.S. [in Ukrainian].
Agarwal, P., & Skupin, A. (Eds.) (2008). Self-Organising Maps: Applications in Geographic Information Science. Wiley, 2008 [in English].
Akinduko, A. A., & Mirkes, E. M. (2012). Initialization of Self-Organizing Maps: Principal Components Versus Random Initialization. ArXiv. https://www.researchgate.net/publication/232503593_Initialization_of_Self-Organizing_Maps_Principal_Components_VersusRandom_Initialization_A_Case_Study [in English].
Bação, F., Lobo, V., & Painho, M. (2005). The self-organizing map, the Geo-SOM, and relevant variants for geosciences. Computers & Geosciences, 31 (2), 155-163. https://doi.org/10.1016/j.cageo.2004.06.013 [in English].
Bianchi, D, Calogero, R., & Tirozzi, B. (2007). Kohonen neural networks and genetic classification. Mathematical and Computer Modelling, 45, 34. https://doi.org/10.48550/arXiv.0809.4755 [in English].
Cottrell, M., Olteanu, M., Rossi, F., & Villa-Vialaneix, N. (2018). Self-OrganizingMaps, theory and applications. ResearchGate. https://www.researchgate.net/publication/325357589_Self-OrganizingMaps_theory_and_applications [in English].
de Bodt, E., Cottrell, M., & Verleysen, M. (2002). Statistical tools to assess the reliability of self-organizing maps. Neural Networks, 15 (8-9), 967-978. https://doi.org/10.1016/S0893-6080(02)00071-0 [in English].
Kohonen, T. (2001). Self-Organizing Maps (3rd ed.). Springer-Verlag Berlin Heidelberg. DOI: https://doi.org/10.1007/978-3-642-56927-2 [in English].
Kohonen, T., & Honkela, T. (2007). Kohonen network. Scholarpedia, 2 (1):1568. https://www.scholarpedia.org/article/Kohonen_network [in English].
Miljković, D. (2017). Brief Review of Self-Organizing Maps. MIPRO. https://www.researchgate.net/publication/317339061_Brief_Review_of_Self-Organizing_Maps [in English].
Umut, A., & Secil, E. (2012). An Introduction to Self-Organizing Maps. In book: Computational Intelligence Systems in Industrial Engineering: with Recent Theory and Applications, (14), 299-319. Publisher: Atlantis Press. https://www.researchgate.net/publication/263084866_An_Introduction_to_Self-Organizing_Maps [in English].
Vayssieres, M. (2024). Master Kohonen Self-Organizing Maps: A Hands-On Guide to Data Exploration with Python. Medium. https://medium.com/@MahounaVAYSSIERES/master-kohonen-self-organizing-maps-a-hands-on-guide-to-data-exploration-withpython-fb92f8ebd6f6 [in English].
Wang, S., Huang, X., Liu, P., Zhang, M., Biljecki, F., etc. (2024). Mapping the landscape and roadmap of geospatial artificial intelligence (GeoAI) in quantitative human geography: An extensive systematic review. International Journal of Applied Earth Observation and Geoinformation, 128. https://doi.org/10.1016/j.jag.2024.103734 [in English].
This work is licensed under a Creative Commons Attribution 4.0 International License.
