Who forecasted it better? The battle between human cops and machine learning to predict future offending

Geoffrey C. Barnes, Ph.D.

Director of Criminology, Capability and Coordination Portfolio, Western Australia Police Force; Vice President, Australia & New Zealand Society of Evidence Based Policing

Advanced statistical techniques are getting better with every passing month, and can be enormously accurate in predicting some kinds of future events.  While these efforts aren’t very common in the criminal justice system – at least not yet – they bring with them an enormous amount of controversy.  Are these statistical methods really all that accurate?  Are they ethical?  Do they mirror the same biases that already exist in the criminal justice system?  Should criminal justice systems be based upon what we think offenders will do in the future, or should we focus only on what they have done in the past?  Previous research has already examined the base levels of accuracy that machine learning can produce when forecasting future offending.  But until now, we haven’t been able to compare these results to the kinds of predictions that ordinary police officers have been making (often quite informally) for the last hundred years.  This paper will examine several different predictive analytics models from a variety of policing jurisdictions, and examine how these forecasts differ from the predictions made by human police officers for the exact same cases.  Who is better at identifying future criminal behaviour?  Man or machine?  This paper will attempt to find out.


Dr Geoffrey Barnes is an Affiliated Lecturer in Evidence Based Policing, supervising students in the Police Executive Programme who are seeking their M.St. in Applied Criminology and Police Management. He has both led and participated in multiple randomised controlled trials, while also performing work on the actuarial forecasting of future criminal behavior, the development of crime and anti-social behaviour over the life course, and the use of cost incentives to promote better outcomes for children in foster care. His research interests also include the use of restorative justice and cognitive behavioural therapy with criminal offenders, the effects of swift and certain sanctions on illegal behaviour, the connections between criminal justice involvement and mortality, and the employment of large data sets derived from official government systems.

He earned his Ph.D. in Criminology from the University of Maryland, and was elected a Fellow of the Academy of Experimental Criminology in 2011. Prior to joining the faculty at Cambridge, he had previous appointments at the University of Pennsylvania, University of Pittsburgh Medical Center, University of Maryland, and Australian National University.

Australia & New Zealand Society of Evidence Based Policing

The Australia & New Zealand Society of Evidence Based Policing (ANZSEBP) was formed in April 2013 in Brisbane, Australia. The ANZSEBP is a police practitioner-led Society. The mission of the ANZSEBP is to develop, disseminate and advocate for police to use scientific research (“the evidence”) to guide best practice in all aspects of policing. The ANZSEBP Chairperson serves on the Executive Board of the British Society of Evidence Based Policing, ensuring that the ANZSEBP works cooperatively with an international group of police to advance evidence based policing.

Conference Managers

Please contact the team at Conference Design with any questions regarding the conference.