Some critiques of KDE challenge the fact that its treats discrete
events as a continuous surface. However this paper is also
concerned with the spread of risk, the risk of having an accident,
geographically will occur not just at a single point but over
a given area. KDE offers a method which takes into account this
notion of spread of accident risk. However one main drawback
reoccurs, which relates to determining the statistical significance
of the resulting clusters. This is an area of research which is something
to investigate in further studies. This study represents a large
scale model for road accident clustering. Further study needs to be
conducted in a number of areas. Firstly, there needs to be development
of amethod for testing for statistical significance of the kernel
density output. Secondly, there needs to be investigation into the
changing dynamics of the clusters over different temporal and spatial
scales. Thirdly, a policy led investigation needs to be conducted
into the suggestions made from the cluster outcomes.