Space Time Lab News Publication
- SpaceTimeLab Launch Event
- Public Crime Mapping Survey
- SpaceTimeLab at Geocomputation 2013
- Congratulations to Berk Anbaroglu and Ed Manley
- SpaceTimeLab welcomes new members
- Job Opportunity at SpaceTimeLab
- Congratulations to James Haworth
- Geospatial Seminar Series - Monsuru Adepeju
- Geospatial Seminar Series - Visualisation of traffic in space-time
- SpaceTimeLab Director Prof. Tao Cheng talks at ASU
- GIS Ostrava 2014
- Geospatial Seminar Series - Prof. Bin Jiang to talk at SpaceTimeLab
- Congratulations to Adel Bolbol
- SpaceTimeLab travels to AAG
- SpaceTimeLab researcher Sarah Wise beats competition to win best paper award at GISRUK 2014
- PhD Studenship Vacancy in Orbital Debris Modelling
- SpaceTimeLab for Big Data Analytics Newsletter 2014
- SpaceTimeLab research top of the pops!
- Juntao Lai wins best paper by a young researcher at GISRUK 2015
- Prof. John Shi on Spatial Data Science
- Prof. Tao Cheng at Evolving GIScience
- SpaceTimeLab Leaflet 2016
- SpaceTimeLab welcomes visitors from Didi and the Shanghai Government
SpaceTimeLab travels to AAG
24 April 2014
In early April, SpaceTimeLab members Sarah Wise, James Haworth and Monsuru Adepeju travelled with SpaceTimeLab Director Tao Cheng to Tampa, Florida, for the Annual Meeting of the Association of American Geographers. The Lab had a strong presence in the Spatiotemporal Thinking, Computing and Applications stream, with presentations on space-time scan statistics (Monsuru), agent-based modelling (Sarah) and machine learning (James), all with application to predictive policing.
In recognition of her standing in the field, Tao Cheng was invited to speak on a panel intended to shape the research agenda for spatiotemporal computing. The session sparked lively debate, leading to a general consensus that we still have a long way to go to truly integrate space and time in our methods and our thinking. Tao Cheng had a busy conference, also presenting research she carried out with former student Tom Wicks on the detection of disaster events from geolocated tweets. The abstracts of each of the talks can be seen below.
How Twitter Reflects Disaster Events - Tao Cheng, Tom Wicks
The real-time nature of citizen journalism makes Twitter a
useful resource for providing information during disaster situations. People
who witness a disaster occurring often turn to Twitter as a service for
information dissemination. They provide data which, when mapped within a GIS
environment, reveals the spatial context of the situation. If used correctly,
this can aid decision makers and help provide assistance to those in need. Here
space-time scan statistics are used to find clusters of tweets across both
space and time. It is expected that clusters will emerge during
spatio-temporally relevant events, as people will tweet more than expected in
order to describe the event to their followers and spread information. Disaster
events are taken as an example of a space-time event, with the 2013 London helicopter
crash used as a case study disaster event.
By using space-time scan statistics, disaster events do lead to statistically significant space-time clusters. In doing this, not only will the effectiveness of using scan statistics on Twitter data be assessed, but also how Twitter reflects disaster events. Each spatio-temporally significant cluster is classified into whether it contains an event, and if so shall be further explored, using key word and network analysis. Knowing how Twitter responds to disasters such as the case presented means tweets can be incorporated into existing decision-making infrastructure to help identify and plan for future potential disaster events.
Cops and Robbers: an Agent Based Model of the Interaction between Policing and Reported Crime Rates – Sarah Wise
Crime science seeks to explain the emergence of crime from the interactions of individuals with their environment and one another. If these patterns of offending are well understood, they can be interrupted or at least mitigated, and their harm to individuals or property can be minimized. However, capturing the dynamics of the complex interplay of processes like opportunistic crime and policing is a challenge for many methodologies. Many kinds of analysis have trouble incorporating features of human decision-making like perception, location, and risk tolerance, all of which are important to the individual's choice whether or not to commit a crime. In light of these difficulties, agent-based modeling can help explore these processes, simulating the behavior of potential offenders as well as that of police officers, highlighting the interactions between these two groups. Drawing upon routine activities theory, essentially rational individuals situated within realistic spatial environments can choose whether or not to commit crimes and pursue the targets they encounter, generating emergent patterns of crime. Their choices are shaped by the perceived gain and their perceived risk, allowing for these essentially human aspects of decision-making to be captured by the simulation. By incorporating the activities of police officers, the possibility of the displacement of crime to other areas or the prevention of repeat victimization can be explored. In the end, the emergent trends are compared with real-world reported crime data.
Machine learning approaches for predictive policing – James Haworth
By their nature, many crime types show patterns of
clustering in time and space. These patterns can be linked to criminological
theories; namely repeat victimisation, routine activity and rational choice.
Such clustering behaviour makes crime forecasting a realistic goal. The
collection of crime data with increasing spatial and temporal resolutions,
coupled with an increasing acceptance of the validity of crime prediction
within law enforcement agencies, has led to the emergence of a new field of
This talk discusses recent developments in crime forecasting models that have emerged from the Crime, Policing and Citizenship project at UCL. Using an extensive dataset of crime data collected in London, this talk first discusses the use of statistical and machine learning approaches for mining patterns in large crime datasets. These patterns are used to forecast the likely future locations of crime.
Detection of Space-Time Crime Patterns for Predictive Policing – Monsuru Adepeju, Tao Cheng, Tomoki Nakaya
Predicting where and when crime is likely to occur is an important aspect of police operation, especially toward the end of effective deployment of limited police resources. Crime hotspot detection is generally used, relying on historical crime datasets, to produce patterns of crime over large temporal steps such as monthly or seasonally. These patterns are then used to project the locations where crimes are likely to occur in near future at the same or larger scales. However, the use of large temporal scales is not suitable for proactive day-to-day crime intervention, especially at the local level. In this study, we combine both the global and local approach of hotspots detection to proactively identify areas of consistently high crime concentration at a finer temporal scale, i.e. daily, over a certain period of time. The results obtained for the two case studies (London - UK & Osaka - Japan) are compared. This approach provides better visualisation and profiling of crime patterns for better predictive policing.
Page last modified on 24 apr 14 11:30