Posts Tagged ‘Shekhar’

Mining for spatial gold

Thursday, February 14th, 2013

Shekhar et al. describe spatial data mining—the process of finding notable patterns in spatial data—and they outline models of doing so, as well as using spatial outliers and spatial co-location rules, and locating spatial clusters. The article is mostly informative, and the topic is central to spatial analysis, so it is difficult to separate spatial mining from the rest of GIS.

I find the notion of clustering particularly interesting, since it is perhaps the most visual-oriented aspect of spatial mining, yet it is largely up for interpretation and/or dependent on the variability of clustering models. For instance, when we see a distribution of points on a map, subconsciously, we begin to see clusters, even if the data is “random.” This type of cognitive clustering is difficult, or even impossible, to model, and it might vary from person to person. The authors of this article list four categories of clustering algorithms, including hierarchical, partitional, density-based, and grid-based models, depending on the order and method of dividing the data. However, the authors fail to note the applications for the various algorithms. If we are thus to naively understand these to be interchangeable, then the results could differ tremendously. Moreover, if there are indeed patterns, then there is most likely a driving force behind those patterns. That force, not the clusters themselves, is the most important discovery in spatial mining and so the modeling must be more stringent in its pursuit of accuracy.

– JMonterey