Development of a robbery prediction model for the City of Tshwane Metropolitan Municipality

Abstract

Crime is not spread evenly over space or time. This suggests that offenders favour certain areas and/or certain times. People base their daily activities on this notion and make decisions to avoid certain areas or feel the need to be more alert in some places rather than others. Even when making choices of where to stay, shop, and go to school, people take into account how safe they feel in those places. Crime in relation to space and time has been studied over several centuries; however, the era of the computer has brought new insight to this field. Indeed, computing technology and in particular geographic information systems (GIS) and crime mapping software, has increased the interest in explaining criminal activities. It is the ability to combine the type, time and spatial occurrences of crime events that makes the use of these computing technologies attractive to crime analysts. This current study predicts robbery crime events in the City of Tshwane Metropolitan Municipality. By combining GIS and statistical models, a proposed method was developed to predict future robbery hotspots. More specifically, a robbery probability model was developed for the City of Tshwane Metropolitan Municipality based on robbery events that occurred during 2006 and this model is evaluated using actual robbery events that occurred in the 2007. This novel model was based on the social disorganisation, routine activity, crime pattern and temporal constraint crime theories. The efficacy of the model was tested by comparing it to a traditional hotspot model. The robbery prediction model was developed using both built and social environmental features. Features in the built environment were divided into two main groups: facilities and commuter nodes. The facilities used in the current study included cadastre parks, clothing stores, convenience stores, education facilities, fast food outlets, filling stations, office parks and blocks, general stores, restaurants, shopping centres and supermarkets. The key commuter nodes consisted of highway nodes, main road nodes and railway stations. The social environment was built using demographics obtained from the 2001 census data. The selection of these features that may impact the occurrence of robbery was guided by spatial crime theories housed within the school of environmental criminology. Theories in this discipline argue that neighbourhoods experiencing social disorganisation are more prone to crime, while different facilities act as crime attractors or generators. Some theories also include a time element suggesting that criminals are constrained by time, leaving little time to explore areas far from commuting nodes. The current study combines these theories using GIS and statistics. A programmatic approach in R was used to create kernel density estimations (hotspots), select relevant features, compute regression models with the use of the caret and mlr packages and predict crime hotspots. R was further used for the majority of spatial queries and analyses. The outcome consisted of various hotspot raster layers predicting future robbery occurrences. The accuracy of the model was tested using 2007 robbery events. Therefore, this current study not only provides a novel statistical predictive model but also showcases R’s spatial capabilities. The current study found strong supporting evidence for the routine activity and crime pattern theory in that robberies tended to cluster around facilities within the city of Tshwane, South Africa. The findings also show a strong spatial association between robberies and neighbourhoods that experience high social disorganisation. Support was also found for the time constraint theory in that a large portion of robberies occur in the immediate vicinity of highway nodes, main road nodes and railway stations. When tested against the traditional hotspot model the robbery probability model was found slightly less effective in predicting future events. However, the current study showcases the effectiveness of the robbery probability model which can be improved upon and used in future studies to determine the effect that future urban development will have on crime.