Mining Spatial Data Repositories: Lessons Learned
Free (open access)
J Pretorius, J Schoeman, H L Viktor & C Blyth
Spatial data mining refers to the extraction of implicit, nontrivial and previously unknown knowledge, in the form of spatial and non-spatial patterns, from spatial data repositories. These repositories include a number space-related data, such as maps, pre-processed remote sensing or medical imaging data, amongst others. In addition, the data repositories are usually ordered using complex spatial indexing structures. Spatial data mining is extremely resource intensive and complex, since it usually involves the mining of huge amounts of spatial data from various diverse applications. The combination of diverse spatial data repositories into an integrated, subject oriented data warehouse aids the data mining practitioner to consolidate various data sources, thus potentially leading to a high quality repository to be used for further exploration. This paper describes the lessons learned during a data mining research effort concerning a census spatial data warehouse. The census data repository concerned the 1996 census of the Greater Pretoria Metropolitan Council area located in the Greater Province of South Africa. Approximately 3 million people, residing in formal or informal settlement within rural and urban areas, inhabit this region. The aim of this research project, which was completed in two steps, was to investigate how spatial data mining may be used to discover new knowledge.. The lessons, as learned through the application of classification and association rule mining, are discussed.