In their article “Spatial Data Mining: A Database Approach” (1997), Ester et al outlined the possibility of knowledge discovery in databases (KDD) using spatial databases, utilizing four algorithms (spatial association, clustering, trends, and classification). Unfortunately, the algorithms are not entirely connected to how one mines spatial information from databases, and the algorithms introduced don’t seem incredibly groundbreaking 20 years later. This paper seemed very dated, particularly because I feel like most of these algorithms are now tools in ESRI’s ArcGIS and the frameworks behind GeoDa, and because the processing issues that seemed to plague the researchers in the late 1990s are not issues (on the same scale) today.
Also, I found it strange that the paper adopted an incredibly positivist approach, and did not mention anything about how these tools could be applied in real life. They acknowledged this as a point of further research in the conclusion, but weighted it less heavily than the importance of speeding up processing times in ‘90s computing. In their introduction, the authors discuss their rationale for using nodes, edges, and quantifying relationships using Central Place Theory (CPT). However, they do not mention that CPT/theorizing the world as nodes & edges is an incredibly detached idea that 1) cannot describe all places, 2) does not realize that human behaviour is inherently messy and not easily predictable by mathematical models, and 3) only identifies trends and cannot be used to actually explain things, just to identify deviances from the mathematical model. Therefore, not everything can be identified by a relationship that a researcher specifies to scrape data using an inherently flawed model, and therefore there will be inaccuracies. It will be interesting to learn if/how spatial data miners have adapted to this and (hopefully) humanized these processes since 1997.