There is an 80.2% chance of crime in my neighborhood next Thursday?

Photo by Campbell Jensen on Unsplash

There is data and then there are criminals.

But what criminals would not want is to be sitting ducks.

Well, in that case they may need to get data-driven and beat the algorithm that predicts crime in cities.

Here is a very interesting read.

An AI algorithm has been developed that predicts crimes a week ahead

The methodology looks great — it is spatio-temporal in nature, the model uses open, public, credible data, segments a city into 300m blocks and uses past data and crime categories to predict crime.

The issues I see with the model:

• It needs to take into consideration the tech and data savviness of the perpetrators. If they are data-savvy, the authorities are bound to create large unmarked gaps in the field for the criminals to play, while being focused on predicted hotspots.
• It needs to consider socio-demographic trends, such as the movement of new people into existing suburbia or nomadic, short-term, floating populations.
• More focus on motive, as a feature
• Human behavior is not predictable. The need for theft or burglary may be driven not be pathological behavior, but by a more imminent need to pay back a much-delayed loan from a loan shark; medicines for a severely ill loved one, or the psychedelic rush of the next drug hit.