Before stepping down as Chief Commissioner of Victoria Police, Ken Lay urged a move away from traditional policing to more sophisticated approaches. With the changing lifestyles and demographics of Victorians and the rapid growth of high-density dwellings, such a move is critical. Victoria’s love affair with high-rise apartment living is creating new challenges for law enforcers with a staggering 63 per cent of property related offences recorded by Victoria Police in 2014.
So, how can homebuyers and renters be sure their property is safe? “Security systems have become a very important and integral part of both residential and commercial living,” says Andrea Baratta, Managing Director at Epsilon Security.
“We have identified a gap in how the security of these buildings is managed,” Andrea says. “The main concern is the lack of a standard method to evaluate security levels of buildings for comparisons. To do this we need to develop a data- driven computer model to assess the security risk and potential crime exposure of a building.”
Andrea proposes a global risk measure be given to all new buildings. He believes this will provide a good starting point for residents seeking to understand how their property compares with others, as well as identify key risks and implement safety improvement solutions.
Epsilon Security accessed funding through an initiative by veski and the Victorian State Government to support Victorian female honours and masters STEM students through a short-term (4 month) tightly focused research internship.
The initiative was delivered by the AMSI Intern program and, in a pilot project, Epsilon Security has teamed up with statisticians from the University of Melbourne to develop modelling to assess property risk and identify safety solutions. Using Victorian crime data the team has been able to investigate demographic / geographical and other measurable variables related to communication, access and monitoring to evaluate the safety of apartment complexes.
“By looking at how a building is accessed, how it is managed, its existing protective measures and what suburb it is located in we can build a computer model to identify weaknesses in buildings and then give them comparative security scores,” says Dr Davide Ferrari from The University of Melbourne’s School of Mathematics and Statistics.
Working with master’s student Puxue Qiao, Dr Ferrari has taken a step closer to achieving this after building a generalised mixed-effect model. These are useful in the social sciences because they are able to take into account both random and fixed effects. For this particular study, where measurements were to be made on clusters of related statistical units repeatedly,choosing a mixed-effect model was natural. Dr Ferrai explains that this type of model allowed them to make predictions of crime occurrences while also obtaining security ratings for each individual building.
“The data collection was done using an original survey through telephone interviews. In the analysis of the frequency of crime occurrences, we used Poisson regression models. A large proportion (36 per cent) of responses obtained for crime occurrences were concentrated at zero. To avoid failure accounting for these zeroes, the team adopted a zero-inflated Poisson model. This model allows for not only a discrete random variable with a Poisson distribution, but also a point mass at zero. We carried our numerical computations of fitting this mixed model using the R package glmmADMB,” says Dr Ferrari.
Giovanni States, when creating a model to assign risk ratings to apartment buildings many variables must be taken into account. To ensure the model was sound, independent legal researcher Dr Giovanni Di Lieto came on board to assist with merging crime prevention literature from criminology in with the statistics. The big challenge, according to Dr Di Lieto, lay in exchanging the different research paradigms. “When we managed to blend our approaches, the significance of this project became not only its substance, but also its methodology. We are able to demonstrate how to effectively interlink industry and different domains of research.”
Qiao described the number of crime occurrences by using a zero-inflated Poisson regression model with mixed effects. A combination of predictors were selected using information theoretical criteria for model selection (Akaike and Bayesian information criteria). The prediction accuracy for the best two selected models were then assessed using independent samples by cross-validation techniques. Using the selected models, Dr Ferrari and Qiao first developed classification methods for the crime risk rating based on the Poisson mixture approach. This offered the most accuracy when predicting the response due to the complexity of the fixed and variable factors.
“The inspiring women industry internship enabled us to collaborate with Dr Ferrari and masters student Qiao to devise a statistically sound model for this project. Qiao also gained valuable industry experience – she was able to solve problems and produce tangible results that have an actual business context,” says Andrea. “Something not nearly enough Australian students have the opportunity to do.”
Universities are drivers of innovation; and they play a key role in projects like this. Given the project’s initial success, the team is applying for an ARC Linkage Project grant to continue their work with the objective of developing a fully automated tool for risk ratings in buildings.