A joint investigation by The Markup and Wired has revealed that Geolitica, a predictive policing software used by the Plainfield Police Department in New Jersey, has been highly ineffective at predicting crimes. Geolitica, formerly known as PredPol, promised to help law enforcement agencies fight crime before it occurs using machine learning algorithms. However, the investigation found that Geolitica’s success rate in accurately predicting crimes was less than half a percent.
The investigation analyzed 23,631 predictions made by Geolitica between February and December 2018. Out of these predictions, less than 100 aligned with actual crimes, indicating a success rate of less than half a percent. While the software performed slightly better in predicting robberies or aggravated assaults (0.6% accuracy) compared to burglaries (0.1% accuracy), the overall results demonstrate the tool’s ineffectiveness.
Captain David Guarino of the Plainfield Police Department admitted that Geolitica was rarely used and questioned its effectiveness in reducing crime. He suggested that the funds spent on the software could have been better allocated to community programs. Geolitica is set to go out of business at the end of the year, with its customers being transferred to SoundThinking, formerly known as ShotSpotter.