South Africa should adopt artificial intelligence since it is used for predictive policing in the US and the UK.

Steven Spielberg’s 2002 film Minority Report (based on a short story by Philip K Dick) imagined a future in which three psychics can “see” murders before they occur. Their clairvoyance enables Tom Cruise and his “Precrime” police force to prevent almost all homicides.

After twenty years, scientists and law enforcement agencies are using data mining and machine learning to mimic those psychics in the real world. This type of “predictive policing” is based on the fact that many crimes – and criminals – have detectable patterns.

 

There have been some successes with predictive policing. In a case study conducted in the United States, one police department was able to reduce gun incidents by 47% during the traditionally gun-crazy New Year’s Eve. In the first 10 weeks of implementing predictive measures, Manchester police were able to predict and reduce robberies, burglaries, and thefts from motor vehicles by double digits.

Predictive policing has advanced dramatically. Previously, humans had to manually sift through crime reports or search through national crime databases. That process can now be automated in the age of big data, data mining, and powerful computers.

 

 

However, simply gathering information is insufficient to deter crime. To detect underlying patterns and relationships, the data must be analyzed. To extract useful information and insights from existing data, scientists use algorithms and mathematical models such as machine learning, which mimics how humans learn.

To refine our approach, we recently turned to an 18th-century mathematical method. We significantly improved crime prediction rates by modifying an existing algorithm based on this method.

This discovery holds promise for using predictive policing in under-resourced settings such as South Africa. This could help reduce crime rates, which are among the highest in the world and are rising. It’s a situation that the country’s police force appears unprepared to handle.

Combining two distinct approaches

Thomas Bayes was a mathematician from the United Kingdom. His famous Bayes’ theorem describes the likelihood of an event occurring according to a prior understanding of conditions that may be related to that event. Bayesian analysis is now widely used in fields ranging from artificial intelligence to astrophysics, finance, gambling, and weather forecasting. We refined the Nave Bayes algorithm and tested it as a crime predictor.

 

Bayesian analysis can use probability statements to answer research questions about statistical models’ unknown parameters. For example, what is the likelihood that a suspect charged with a crime is guilty? Going deeper, however, requires increasingly sophisticated technologies and algorithms, such as calculating how poker cards may unfold or how humans (especially humans with criminal intent) will act.

For crime prediction, we used the Nave Bayes algorithm or classifier, a popular autonomous machine learning algorithm.

The premise of Nave Bayes is that features – the variables that serve as input – are conditionally independent, which means that the presence of one feature has no effect on the others.

 

We improved the Nave Bayes algorithm by combining it with another known as Recursive Feature Elimination. This tool assists in selecting the more significant features in a dataset and removing the less significant ones in order to improve the results.

We then applied our refined algorithm to a well-known experimental dataset extracted from the Chicago Police Department’s CLEAR (Citizen Law Enforcement Analysis and Reporting) system, which has been used to predict and reduce crime in that American city. Because of the rich data contained in that dataset, it has been used globally: it provides incident-level crime data, registered offenders, community concerns, and the locations of police stations in the city.

 

We compared the results of our improved Nave Bayes to those of the original Nave Bayes, as well as those of other predictive algorithms like Random Forests and Extremely Randomized Trees (algorithms we have also worked on for crime prediction). We discovered that we could improve on the Nave Bayes predictions by about 30%, and that we could either match or improve on the predictions of the other algorithms.

Information and bias

While our model has promise, there is one critical component missing when applying it to South African contexts: data. Predictive models work best when there is a large amount of relevant data to work with, as demonstrated by the Chicago CLEAR system. However, due to confidentiality concerns, South Africa’s police force has historically been very tight-fisted with its data. In my doctoral research on detecting and mapping crime, I ran into this issue.

This is gradually changing. We are currently conducting a small case study in Bellville, a suburb about 20 kilometers from Cape Town’s central business district and the location of our university, using South African Police Service data for predictive policing.

 

None of this is to say that predictive policing will solve South Africa’s crime problem on its own. Both predictive algorithms and policing have flaws. Even the psychics in Minority Report were not without flaws. Concerns have been raised in South Africa and elsewhere that these algorithms may simply reinforce racial biases.

However, we believe that, with ongoing technological advancement, predictive policing could play an important role in improving police responsiveness and may be a small step toward improving public trust in the police.