Crime incidence currently has annual variations at monthly resolution (this is all we get from the open data)
But crime occurrence times will be heavily dependent on crime type .e.g. shoplifting only happens when shops are open, assaults more common on Friday nights etc.
Propose that a day is split into 3 segments: night, day, evening thus creating 21 buckets per week. The current flat (monthly) crime incidences can then be given a weekly periodicity (preserving the mean) and sampled on a more frequent basis than annually (as in done now)
Crime incidence currently has annual variations at monthly resolution (this is all we get from the open data)
But crime occurrence times will be heavily dependent on crime type .e.g. shoplifting only happens when shops are open, assaults more common on Friday nights etc.
Propose that a day is split into 3 segments: night, day, evening thus creating 21 buckets per week. The current flat (monthly) crime incidences can then be given a weekly periodicity (preserving the mean) and sampled on a more frequent basis than annually (as in done now)