In this paper we demonstrate
the use of network analysis to characterise patterns of clustering in
spatio-temporal events. Such clustering is of both theoretical and practical
importance in the study of crime, and forms the basis for a number of
preventative strategies. However, existing analytical methods show only that
clustering is present in data, while offering little insight into the nature of
the patterns present.
Here, we show how the classification of pairs of events
as close in space and time can be used to define a network, thereby
generalising previous approaches. The application of graph-theoretic techniques
to these networks can then offer significantly deeper insight into the
structure of the data than previously possible.
In particular, we focus on the
identification of network motifs, which have clear interpretation in terms of
spatio-temporal behaviour. Statistical analysis is complicated by the nature of
the underlying data, and we provide a method by which appropriate randomised
graphs can be generated. Two datasets are used as case studies: maritime piracy
at the global scale, and residential burglary in an urban area.
In both cases,
the same significant 3-vertex motif is found; this result suggests that
incidents tend to occur not just in pairs, but in fact in larger groups within
a restricted spatio-temporal domain. In the 4-vertex case, different motifs are
found to be significant in each case, suggesting that this technique is capable
of discriminating between clustering patterns at a finer granularity than
previously possible.
Below: The mapping between event sets and their event networks. This schematic diagram represents the mapping between the
space of event-sets and the space of event networks, for an illustrative case
involving 4 events. For simplicity, the events are all assumed to be close
temporal pairs, and the region is taken to be a square with width equal to the
spatial close pair threshold. Only one configuration—that in which the events
occur at the extremes of the region—maps to the empty network, whereas many
possible event-sets map to the fully-connected network.
Full article at: http://goo.gl/Pa7Np3
By: Davies T1,2, Marchione E2,3.
- 1Department of Civil, Environmental and Geomatic Engineering, University College London, London, United Kingdom.
- 2Department of Security and Crime Science, University College London, London, United Kingdom.
- 3Centre for Advanced Spatial Analysis, University College London, London, United Kingdom.
More at: https://twitter.com/hiv_insight
No comments:
Post a Comment