Issues with Unsupervised Classification in Spatial Databases
Geography and Geology
This study evaluated a spatial database for Long Beach, Mississippi, using Geographic Information System (GIS) software provided by the Environmental Research System Institute (ESRI) software, ArcINFO 10.0. The data sample came from a section of spatial data funded by the State of Mississippi that has proven to be problematic for emergency responders in Long Beach. The spatial database is a vector data building footprint inventory of Harrison County and a point data address point tied to the building footprint centroid. We found a coverage gap of 19.8% of the buildings and 35.7% of the parcels in the approved and produced dataset. It is our opinion that supervised classification is a more accurate method of producing datasets over unsupervised classification. Unsupervised classification is useful for gap analysis, data foundation and detecting change, but supervised classification of discrete entities like building footprints is needed to produce high levels of data integrity and reliability useful for emergency responders. Furthermore, an information technology specialist needs to be involved in the final approval of spatial datasets to ensure data quality.
International Journal of Applied Research on Information Technology and Computing
Holt, D. H.,
(2012). Issues with Unsupervised Classification in Spatial Databases. International Journal of Applied Research on Information Technology and Computing, 3(3), 157-161.
Available at: https://aquila.usm.edu/fac_pubs/15643