A multiple window-based co-location pattern mining approach for various types of spatial data

Studies on spatial co-location mining required distance threshold to define spatial neighbourhood (Shashi Shekhar and Yan Huang(2001); Yoo and Shekhar (2004, 2006); Yasuhiko Morimoto(2001); Koperski and Han(1995); Ding et al. (2008)) However, it is problematical for users to choose suitable threshold values because they lack prior knowledge about spatial data. Spatial neighbourhood has been defined generally by various methods namely distance, direction and/or topology (Shashi Shekhar and Sanjay Chawla 2003). In addition, spatial data is not usually evenly distributed, so a unique distance value cannot fit all spatial datasets well. Hence, the result of co-location pattern mining depends on the definition of neighbourhood to describe the mutual relationship of spatial objects. Discovering a co-location pattern from extended spatial objects like line and polygon is another challenge. Handling line and polygon spatial data are much more complex than point spatial data. Modelling and structuring are inter-related and play vital role in the success of informative mining of spatial data This chapter discusses the usage of an event centric approach and N-most prevalent approach for analysis of co-location patterns. Both the approaches have used distance threshold to define spatial neighbourhood. The window based model has been proposed to find the neighbourhood for point spatial data sets and the multiple window model has been proposed for extended spatial data objects. Generic and efficient multiple window based model algorithm has been proposed for co-location patterns. Finally, the proposed approach has been compared with existing approaches.