A computer system was able to predict in advance and with great accuracy the crime rates in two US cities
Uriel Brizon
Direct link to this page: https://www.hayadan.org.il/simpesha.html
A computerized system developed in the USA will allow the police to predict in advance which urban areas crimes will be committed and to allocate resources to these areas ahead of time. The system was developed by a group of scientists from Carnegie Mellon University in Pennsylvania, led by Prof. Wilfen Gore, and it predicts activity in urban crime centers based on analysis Patterns of events from the past were analyzed by the system in a recent test A large crime database, containing records of about six years of police activity, the system showed impressive predictive ability and the research team announced that it could become operational within a year.
If it is proven that the system accurately predicts crime trends even in real field conditions, police forces that purchase it will be able to predict in advance where and with what intensity urban crime will strike. This is similar, in a certain sense, to weather forecasts that make it possible to predict rainstorms in advance and prepare for them.
To make the predictions, the system uses advanced statistical techniques and neural networks. Maps indicating areas according to the frequency of crime incidents have been in use for a long time, but only recently have researchers begun to use the large databases accumulated on crime incidents to develop tools for more accurately predicting future crime events.
In the database used by the team of researchers, information items were stored on approximately 6 million crime incidents that occurred in recent years in two US cities: Pittsburgh in Pennsylvania and Rochester in the state of New York. The large amount of information accumulated allowed the researchers to use computer prediction methods that had not been tried before.
The scientists used a two-step technique to analyze the data. In the first stage, they performed a statistical analysis of the data with the aim of uncovering the basic laws of behavior that characterize urban crime and formulating them in a way that would allow for future prediction. The research team looked for data of high importance in the forecasting process (indicators) and laws that describe well the trend changes observed in the data. An indicator can be given
individual, such as the average time between two crime incidents, and a law can be, for example, a description of the relationship between two different types of crime. During the analysis, it became clear to the researchers that a large part of the information they collected basically describes, in a quantitative form, knowledge that police officers naturally accumulate during the years of their duties.
The "laws of the field" of urban crime, known to every experienced police officer, were recorded for the first time in an orderly and systematic manner. Thus, for example, the trend of the increase in shoplifting before the Christmas season was measured and the connection between the number of cases of vandalism in a certain area and the trend of an increase in more serious offenses in that area was identified. The researchers prepared a long list of laws and indicators identified as having predictive ability and implemented them in the system.
In the second step, the researchers added a processing layer containing a neural network to the system. Neural networks are computational systems somewhat similar to the brain of an animal. They simulate the operation of large groups of neurons by building many connections between virtual cells in the computer's memory, and are able to change the strength of these connections as needed. The use of neural networks allows computer systems to deal with pattern recognition problems that are particularly difficult for computers. Through the use of the neural network, the researchers gave the system the ability to identify subtle patterns of behavior and small, but important trends that were not detected by the statistical analysis.
The statistical mechanism and the neural network are complementary factors in the forecasting work. While the rules that guide the statistical analysis are fixed, understandable and contain knowledge entered into the system in a controlled manner, the neural network works in a different way. It adapts itself to the data and consolidates a set of knowledge that is not defined in advance but emerges from the data itself. The way computerized neural networks work is somewhat similar to the way animals learn from experience. Success in forecasting strengthens one course of action and weakens another. It is not always necessarily clear what the law underlying the improved predictive ability is, but when success is recorded in the prediction, various internal relationships are strengthened and the law is established in the system. Through a combination of the built-in statistical layer and the dynamic analysis of the neural network, the system was able to make more accurate predictions than before and display the ability to function as an operational tool under real conditions.
In an examination that was based on information accumulated over 72 months on crime data in Pittsburgh and Rochester, the system predicted crime rates with an average deviation of only 20-10%, and in very small urban areas, up to three square kilometers. In cases where there was a sudden change in the central indicators , the accuracy rate of the prediction decreased, but in any case it remained higher than 50% if it was possible to predict the weather in such an area small and at such a level of precision, it was an impressive predictive ability.
After the tests are completed, it will be possible to distribute the system to police stations around the world. According to the team of researchers, no prior knowledge is required to operate it and its use will be possible immediately upon installation. In the discussions conducted on how to operate the system, the interesting question came up: Is it worth publishing certain data on crime forecasts to the general public, just as the weather forecast is published. The publication of the data will, without a doubt, harm the accuracy of the prediction, since it will provide information to the criminals about the preparation of the police. On the other hand, it will be possible to make intelligent use of advertising to intentionally influence crime trends, in a kind of psychological warfare. The researchers estimate that the decision that will be made is to avoid publishing data on the crime forecast even in the cities where the system will be activated.
Imagine your reaction to the announcement in the "Crime Forecast Corner", at the end of the news edition, that a 70% increase in violent crime is expected in your area of residence; The expected public panic in such a case will probably not allow public publication, but for the intelligence departments of police forces around the world, the system may become an irreplaceable tool in the fight against crime.
Knowledgeable in robotics and artificial intelligence
https://www.hayadan.org.il/BuildaGate4/general2/data_card.php?Cat=~~~621497287~~~65&SiteName=hayadan
