August 7, 2012 at 13:01Algorithm Predicts Crime Tracking Mobile Phones
For many years, scientists have been experimenting with algorithms that can predict crime. It is assumed that criminals tend to repeat successful actions - at least they do not use the RNG to select the place and time of the crimes, so their actions are predictable by definition.
For example, a year ago, the Californian city of Santa Cruz was the first in the world to introduce a mathematical model for calculating the probability of crime , which every day forms a new route for patrol cars based on street crime statistics. It takes into account the day of the week, time of day, the presence / absence of football matches on TV and other factors.
Mirco Musolesi, a researcher at the University of Birmingham, took a completely different approach. His method is not based on statistics, but on live data from cellular networks. Musolesi began by teaching the algorithm to predict the movements of each subscriber with a high degree of probability: he even won the Nokia Mobile Data contest, most accurately predicting the movements of 25 volunteers by their phone signals, call history and text messages. Sometimes the algorithm predicts the user's coordinates with an accuracy of 20 m 2 .
The algorithm works efficiently only if the entire network of friends of the specified user is monitored simultaneously. If you track only one person, then the accuracy of the prediction of coordinates is reduced to 1000 m 2 . Even if it was possible to extract clarifying information from just one friend, the accuracy immediately increased sharply.
Thus, the algorithm is able to calculate the place and time where in 20-30 minutes a group of potential criminals will meet. According to Musolesi, you can calculate a specific street, block, sometimes even a house - the scene of a potential crime. Obviously, there just in case you need to send a patrol car.
Musolesi hopes that his development will be used by law enforcement agencies. The researcher is sure that such a data analysis system does not violate the law: “Our algorithm is a way to extract new information from data [which the police already have],” he says in an interviewForbes Edition. Someone would call such a method dubious, because surveillance is carried out for people who have not yet committed a crime, but only by some criteria are listed in the “risk group” (for example, citizens with a criminal past or mentioning a keyword in a telephone conversation or in SMS). On the other hand, the algorithm itself works with anonymous data, so this “tracking” is fundamentally no different from any contextual mobile advertising system or chat monitoring system by keywords, which is used in Facebook chats and other services.
In principle, such algorithms can be used not only by the police, but also ordinary commercial companies. Everyone will be pleased when you go to your favorite cafe - and there they just laid the table for you. Or you approach the house - and the kettle automatically turns on in the kitchen and dinner begins to prepare.
Mirko Musolesi with colleagues published the results of their work (pdf), in the near future they are going to conduct additional experiments on the database provided by Nokia.
For example, a year ago, the Californian city of Santa Cruz was the first in the world to introduce a mathematical model for calculating the probability of crime , which every day forms a new route for patrol cars based on street crime statistics. It takes into account the day of the week, time of day, the presence / absence of football matches on TV and other factors.
Mirco Musolesi, a researcher at the University of Birmingham, took a completely different approach. His method is not based on statistics, but on live data from cellular networks. Musolesi began by teaching the algorithm to predict the movements of each subscriber with a high degree of probability: he even won the Nokia Mobile Data contest, most accurately predicting the movements of 25 volunteers by their phone signals, call history and text messages. Sometimes the algorithm predicts the user's coordinates with an accuracy of 20 m 2 .
The algorithm works efficiently only if the entire network of friends of the specified user is monitored simultaneously. If you track only one person, then the accuracy of the prediction of coordinates is reduced to 1000 m 2 . Even if it was possible to extract clarifying information from just one friend, the accuracy immediately increased sharply.
Thus, the algorithm is able to calculate the place and time where in 20-30 minutes a group of potential criminals will meet. According to Musolesi, you can calculate a specific street, block, sometimes even a house - the scene of a potential crime. Obviously, there just in case you need to send a patrol car.
Musolesi hopes that his development will be used by law enforcement agencies. The researcher is sure that such a data analysis system does not violate the law: “Our algorithm is a way to extract new information from data [which the police already have],” he says in an interviewForbes Edition. Someone would call such a method dubious, because surveillance is carried out for people who have not yet committed a crime, but only by some criteria are listed in the “risk group” (for example, citizens with a criminal past or mentioning a keyword in a telephone conversation or in SMS). On the other hand, the algorithm itself works with anonymous data, so this “tracking” is fundamentally no different from any contextual mobile advertising system or chat monitoring system by keywords, which is used in Facebook chats and other services.
In principle, such algorithms can be used not only by the police, but also ordinary commercial companies. Everyone will be pleased when you go to your favorite cafe - and there they just laid the table for you. Or you approach the house - and the kettle automatically turns on in the kitchen and dinner begins to prepare.
Mirko Musolesi with colleagues published the results of their work (pdf), in the near future they are going to conduct additional experiments on the database provided by Nokia.