Page 11 - Case study: Crime prediction for more agile policing in cities – Rio de Janeiro, Brazil

CrimeRadar applies leading edge mathematics, advanced machine learning tools, and an easy to
use interface to help translate historical crime data into more accessible and actionable information
for users. CrimeRadar is not intended to prescribe or mandate safety levels in a region. Rather, it is
intended to serve as one additional data point that the public can use when making decisions about
traveling to specific locations.

CrimeRadar is an example of public private partnership whereby actual city crime data was provided
to Igarape Institute and its partners to develop a solution which helps both city police as well as
citizens and visitors of the city. The Institute provided expert knowledge about the region and
worked with various data providers to gather and verify accurate historical data. Via Science helped
to create and test the formula behind the app using their proprietary machine learning software
architecture. Finally, Mosaico, a local software firm, helped design the mobile app interface.

2.3. Results

Given its initial success in Rio de Janeiro, starting in 2018, the Institute partnered with the state
military police of Santa Catarina to develop and pilot a police-facing version of CrimeRadar. In 2020,
the initiative will be evaluated using a randomized controlled trial (RCT) to assess the effectiveness
of predictive policing on the planning of police patrol itineraries and scheduled operations. The
intention is to assess changes in crime levels and crime displacement as well as average police
response times and public trust in the police.

More importantly, the entire development process and resulting crime forecasting algorithm are
being documented in a "social impact statement". The intention is to describe the challenges
associated with (and steps taken to ensure) the monitoring and analysis of crime data. The Igarapé
Institute is committed to designing and implementing crime forecasting algorithms in a publicly
accountable manner.

To start, the software license will require that all police departments deploying the predictive
tool comply with a minimum set of transparency and reporting standards. A list of five minimum
requirements are set out below and were prepared by the FAT/ML work group, a community of
8
researchers and practitioners concerned with fairness, accountability, and transparency in machine
learning:

• Responsibility and Recourse – Make available externally visible avenues of redress for adverse
individual or societal effects of an algorithmic prediction system, and designate an internal role
for the person who is responsible for the timely remedy of such issues.

• Explainability – Ensure that algorithmic predictions as well as any data driving those predictions
can be explained to end-users and other stakeholders in non-technical terms.

• Accuracy – Identify, log, and articulate sources of error and uncertainty throughout the data
sources so that expected and worst case implications can be understood and inform mitigation
procedures.

Crime prediction for more agile policing in cities – Rio de Janeiro, Brazil 5