Shekhar et al. (2003) outline various techniques in spatial data mining which can be used to extract patterns from spatial datasets. In discussing techniques for modeling spatial dependency, detecting spatial outliers, identifying spatial colocation, and determining spatial clustering, Shakhar et al. effectively demonstrate the relevant challenges and considerations when working with a spatial dataset. Due to factors such as spatial dependency, and spatial heterogeneity, “general purpose” data mining techniques will perform poorly on spatial datasets and new algorithms must be considered (Shekhar et al., 2003).
Shekhar et al. define a spatial outlier as a “spatially referenced object whose non-spatial attribute values differ significantly from those of other spatially referenced objects in its spatial neighbourhood” (p 8). I have not previously been exposed to research on spatial outliers, but I was surprised to read such a definition in which an outlier is determined by its non-spatial attribute. I am left wondering if it is possible to invert Shekhar’s definition and define spatial outliers in terms of differences in spatial attribute values among objects with consistent non-spatial attribute values. For example, when talking about the locations of bird nests, could we define a spatial outlier as a nest which is significantly far from a cluster of other nests?
As this article was broadly speaking about knowledge discovery from spatial datasets, I was reminded of last week’s lecture on geovisualization. While the objective approach of spatial data mining contrasts the exploratory geovisualization process, I am curious how the two approaches can effectively be combined to drive a more holistic process of knowledge discovery from spatial data.