Don't miss our weekly PhD newsletter | Sign up now Don't miss our weekly PhD newsletter | Sign up now

  Near-real time and microscale crime prediction


   Faculty of Engineering, Computing and the Environment

   Thursday, November 07, 2024  Competition Funded PhD Project (Students Worldwide)

About the Project

Crimes incur significant financial, societal and emotional costs, and their reduction is invariable to improving the safety of our society. Crime prediction has been utilised as a vital element in supporting the policing effort in the fight against crime, and a series of methods have been developed to this end. Broadly speaking, the classic methods mainly rely on descriptive and regressive models that offer strengths in aggregate and long-term projections, while the recent, data-driven methods largely depend on machine-learning (ML) approaches to understand the patterns of crime across space and time for each type of crime to make a prediction of the likely future occurrences. While these approaches are useful in making near-future predictions at various granularity, no de facto standard has been established for making accurate predictions.

Limitations of the existing index of diversity

In recent years, many ML techniques have been adopted for the purpose of crime prediction, ranging from random forests to decision trees, KNN and naïve Bayes. Existing studies (e.g. Alsubayhin 2023) suggest that supervised learning tend to yield more accurate results, while random forests is most commonly used in the context of crime prediction. However, the goodness of fit and the robustness in predicting at different spatial granularity and for different forecast period tend to vary between the crime types, the local geography and even by the seasons. They also rely mainly on the past crime data as the predictor, even though the policing response is becoming increasingly important in affecting the eventual crime outcomes. Rotaru et al. (2022) reports that police responses tend to vary, despite the ML predictions made, and thus significantly affecting the outcomes.

What this project aims to achieve

This project is aims to do two things:

(1) It first expands on the recent development of ML-based crime prediction approaches by evaluating the accuracy in their predicting powers and the applicability of these approaches to a range of contexts (including the variation arising from geographical patterns, weather conditions, and daily, weekly and seasonal oscillation). The goal is to identify whether a single method can be identified for its overall performance, or formulate a framework consisting of a series of methods that can collectively make forecasts with a high level of accuracy.

(2) This is followed by designing an interactive model that predicts the impact of crime through crime prediction and the police responses. Using the existing patterns of crime from the past data as well as the police responses at the time, we build a prediction tool that will project likely scenarios based on the different policing effort to be taken against the predicted future. The police response data is partly available in the form of emergency response calls, and the team is currently also negotiating access to detailed police dispatch data which will give a fuller picture for crime prediction.

Through these two steps, our team aims to establish a framework that offers a robust and near-real-time crime prediction with different scenarios based on the impact of police responses.

 

What we are looking for in the prospective student

We are looking for an ambitious, forward thinking person with background in quantitative geography and/or data science, who has a wide-ranging interest in cross-disciplinary research and can think laterally to compare the likely outcomes under different scenarios and with varying policing responses.


Computer Science (8) Geography (17)

Funding Notes

This project may be eligible for a Faculty studentship - see the information at Research degrees - Faculty of Engineering, Computing and the Environment - Kingston University

Funding available:

Fees: Home tuition fee for 3 years full-time

Stipend: .£21,237 per year for 3 years full-time

International students will be required to pay the difference between the Home and International tuition fee (£13,000 approx for 2024-25)


How to apply: see the Faculty Studentships information at Research degrees - Faculty of Engineering, Computing and the Environment - Kingston University


References

Alsubayhin, A. et al. (2023) Crime prediction using machine learning: A comparative analysis. Journal of Computer Science 19(9): 1170-1179.
Rotaru, V. et al. (2022) Event-level prediction of urban crime reveals a signature of enforcement bias in US cities. Nature Human Behaviour 6: 1056-1068.
Shiode, N., Shiode, S., Nishi, H. & Hino, K. (2023) Seasonal characteristics of crime: An empirical investigation of the temporal fluctuation of the different types of crime in London, Computational Urban Science 3(19).
Shiode, N., Shiode, S. & Inoue, R. (2022) Measuring the colocation of crime hotspots. GeoJournal 88, 3307-3322.
Shiode, S. & Shiode, N. (2020) Street-level crime geo-surveillance in micro-scale urban environment – NetSurveillance, Annals of the American Association of Geographers 110: 1386-1406.
Shiode, S. & Shiode, N. (2022) Network-based space-time Scan Statistics for detecting micro-scale hotspots, Sustainability 14(24), 16902.

Register your interest for this project