An optimised approach to near repeat analysis for intelligence driven crime linkage

ABSTRACT

Law enforcement and security agencies around the globe have integrated geospatial analysis into their intelligence workflow to profile serial offenders, track suspects, and direct crime reduction/prevention efforts. Expansion to spatio-temporal analyses may yield significant and relevant information to better understand the underlying factors of crime. Among the current spatio-temporal methods to associate crimes is near repeat analysis. The premise of the near repeat phenomenon is that if a given location is the target of a crime, nearby locations will have an increased chance of being targeted for a limited time with the level of risk decaying with distance from the original target and over time. Robust analytical methods were developed to discover and further understand spatio-temporal clustering of crime incidents. The open source nature of these functions facilitate transparency and reproducibility in the analytical method and implementation across agencies/police management systems. Firstly, a new method for near repeat analysis is presented which expands current techniques through graphical linkage of crime incidents given spatio-temporal proximity. Next, this method is used to evaluate the prevalence of near repeats across cities of scale. Given this, a method for determining optimal parameters is presented and utilised to determine the optimal parameters (inter-incident time/distance).

KEYWORDS:

• Crime
• space–time clustering
• near repeat
• open source
• network

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Connection to an existing database can be achieved with an ODBC driver, see the odbc package (Hester & Wickham, 2020) for further information.

2 At the highest parameters, inter-incident distance of 5000 meters and inter-incident time of 365 days, all incidents were linked in a single network.

3 The rationale of reducing the computational demand was to ensure that agencies with limited computational capacity could still implement the technique.

4 The interactive nature of these surface plots can be achieved through the near_repeat_analysis and near_repeat_eval functions within the rcrimeanalysis package. Additionally, there are vignettes integrated in the software documentation which provide examples of the analytical methodology.

5 Note that due to data availability, this was not possible for all cities being evaluated.

6 Not all fields/columns of incident information are included in Table 4.

7 Researchers can also utilize the function in conjunction with sample crimes data in the rcrimeanalysis package or with open crime data used in this project. The development version of the package is available in a GitHub repository at: https://github.com/JSSpaulding/rcrimeanalysis.

8 One notable difference is that Baltimore has a higher volume of crime, and the z axis goes far higher.

9 A small distance may also be useful if attempting to detect crimes repeat crimes in apartment buildings or complexes.

10 The example is abbreviated because not all spatial and temporal bandwidths are presented. There were ten spatial and ten temporal bandwidths computed, including a ‘More than’ field to encapsulate all points.