Why it matters: Law breakers have been posting about criminal behavior on social media for as long as the medium has been around but as recent research highlights, police don't need people to directly tattle on themselves to predict where crime will happen.
Law enforcement agencies have established crime prevention techniques dating back decades. Built on metrics like historical patterns of events, geographical information and local demographics, this data helps agencies develop crime prevention tactics that improve patrol strategies, increase public safety and decrease economic loss.
The problem, however, is that these variables change slowly over time and do not capture short-term variances associated with real-life crime events.
Fortunately for law enforcement, they’re able to keep their finger on the pulse of the public through modern technology such as social media.
As examined in a recent study on the subject, social media platforms like Twitter and Foursquare generate massive amounts of data that provide unprecedented opportunities to capture a city’s dynamics. By feeding seemingly unsuspecting data into an algorithm – for example, check-ins from Foursquare – researchers were able to use the routine activity theory to determine that a location with users across a diverse background is likely to spawn certain types of crime such as theft.
In experiments in Brisbane and New York City, researchers noted that Area Under Curve (AUC) values improved after factoring in dynamic features (social media information). Using the Random Forest prediction model, valued increased four percent for theft, four percent for drug offenses, 16 percent for assault, two percent for fraud and six percent for unlawful entry. In New York City, values improved four percent for theft, two percent for drug offenses, two percent for assault, four percent for fraud and four percent for unlawful entry.
In both cases, traffic related offenses weren’t impacted by social media intelligence.
https://www.techspot.com/news/77086-social-media-data-effective-predicting-crime.html
