February CIC Research Seminar

Catching Crooks with Bayes Nets: Using Bayesian Networks to identify significant crime events

Presented by Mr Shih Ching Fu

Abstract:

Police globally have invested heavily in curating accurate crime data on their respective jurisdictions. Anecdotally however, senior officers are spending proportionally more and more time wading through these new lakes of data and proportionally less time making timely policing decisions.

As an example, the Daily Crime Review, delivered to the WA Deputy Commissioner of Police every morning, comprises a 15 minute verbal briefing from his staff officer outlining the significant criminal incidents from the past 24 hours. Incidents mentioned in this briefing are judged significant based on details such as the suspects and victims involved, harm inflicted on officers, and perceived threat to community safety. Currently, the preparation of this Daily Crime Review takes more than 60 minutes; an onerous task since the backgrounds of around 600 criminal incidents must be reviewed, contextualised, and synthesised each morning by hand.

In this talk I’ll describe our Bayesian belief network for modelling crime incident significance, as perceived by the Deputy Commissioner. Our model attempts to encapsulate the intuition of the officers who routinely prepare the Daily Crime Review as they screen incidents; discovering their thought processes through structured and unstructured interviewing and observation. In testing, the resultant model rapidly classifies approximately three quarters of incidents as ‘insignificant’ therefore saving staff officers the effort of delving any deeper into those incidents. Planning is underway to further refine this model and eventually develop an app accessible to officers.

Bio:

Shih Ching is a Masters student in Mathematics at Curtin University who in 2019 undertook a project with WA Police, the subject of this presentation.

With a background in computer science, he’s returned to study statistics after a varied career in academic research, enterprise software development, and product management at a tech start-up. His newly discovered zeal for applied statistics means he’s super keen to learn new techniques and work on cross-disciplinary projects. Bayesian networks are his latest fixation but logistic regression is his first love.

When off campus, his time is normally distributed between his children, wife, and his 3-iron.

The presentation will be followed with a networking opportunity.