In traditional crime prediction models, the use of census data has been limited as it fails to capture the complexity and dynamics of human activity. However, with the advent of ubiquitous computing, there is an opportunity to improve these models by incorporating data that better represent human presence in cities. In their paper titled "Mining large-scale human mobility data for long-term crime prediction," Cristina Kadar and Irena Pletikosa leverage large human mobility data to develop a comprehensive set of features for crime prediction. These features are informed by theories in criminology and urban studies. The authors employ averaging and boosting ensemble techniques from machine learning to investigate the predictive power of these features for different types of crimes occurring at the census tract level in New York City. They find that spatial and spatio-temporal features derived from Foursquare venues and check-ins, subway rides, and taxi rides significantly improve baseline models that rely solely on census and point-of-interest (POI) data. The proposed models achieve impressive absolute R^2 metrics, with up to 65% accuracy on a geographical out-of-sample test set and up to 89% accuracy on a temporal out-of-sample test set. This demonstrates that, in addition to the residential population of an area, the ambient population strongly predicts crime levels in that area. The authors delve deeper into specific crime categories and observe that the predictive gain of human dynamics features varies across different types of crimes. For instance, these features provide the greatest boost in predicting grand larcenies while assaults are already well predicted by census features alone. Furthermore, Kadar and Pletikosa identify and discuss the top predictive features for each main crime category. These findings offer valuable insights for those responsible for urban policy or law enforcement. By leveraging geo-tagged human dynamics data such as public transportation or taxi usage patterns researchers can now empirically measure aspects of criminological theories that were previously difficult to study at a large scale. Overall this research highlights the potential of using large-scale human mobility data for long-term crime prediction models. By incorporating these dynamic data sources such as Foursquare venues and check-ins subway rides and taxi rides alongside traditional census data policymakers and law enforcement agencies can enhance their understanding of crime patterns and develop more effective strategies to ensure public safety and security.
- Traditional crime prediction models have limited use of census data
- Ubiquitous computing provides an opportunity to improve these models by incorporating data that represent human presence in cities
- The paper titled "Mining large-scale human mobility data for long-term crime prediction" by Cristina Kadar and Irena Pletikosa uses large human mobility data to develop features for crime prediction
- Features are informed by theories in criminology and urban studies
- Spatial and spatio-temporal features derived from Foursquare venues, check-ins, subway rides, and taxi rides significantly improve baseline models that rely solely on census and point-of-interest (POI) data
- Proposed models achieve impressive absolute R^2 metrics with high accuracy on geographical and temporal out-of-sample test sets
- Ambient population strongly predicts crime levels in addition to residential population
- Predictive gain of human dynamics features varies across different types of crimes, with the greatest boost seen in predicting grand larcenies
- Top predictive features for each main crime category are identified and discussed
- Geo-tagged human dynamics data can be leveraged to measure aspects of criminological theories at a large scale
- Using large-scale human mobility data alongside traditional census data can enhance understanding of crime patterns and develop more effective strategies for public safety.
Traditional crime prediction models: These are ways of trying to figure out where crimes might happen based on certain information.
Census data: Information collected about people and their living situations, like how many people live in an area or what kind of jobs they have.
Ubiquitous computing: This means using technology that is all around us, like smartphones or sensors, to help make predictions about crimes.
Human mobility data: Information about how people move around in a city, like where they go and how they get there.
Features: These are specific things that can be used to predict something. In this case, features are used to predict crimes.
Criminology: The study of crime and why it happens.
Urban studies: The study of cities and how they work.
Spatio-temporal features: Features that take into account both space (where something is) and time (when something happens).
Foursquare venues: Places that people can go to, like restaurants or stores, that are listed on the Foursquare app or website.
Check-ins: When someone uses a smartphone app or website to say where they are at a certain time.
Subway rides and taxi rides: Using public transportation or taxis to get from one place to another in a city.
Baseline models: Basic models that only use certain types of information to try and predict crimes. In this case, the baseline models only use census data and point-of-interest data (like Foursquare venues).
Absolute R^2 metrics with high
Mining Large-Scale Human Mobility Data for Long-Term Crime Prediction
Crime prediction models have traditionally relied on census data to capture the complexity and dynamics of human activity. However, with the advent of ubiquitous computing, there is an opportunity to improve these models by incorporating data that better represent human presence in cities. In their paper titled "Mining large-scale human mobility data for long-term crime prediction," Cristina Kadar and Irena Pletikosa leverage large human mobility data to develop a comprehensive set of features for crime prediction. These features are informed by theories in criminology and urban studies.
Machine Learning Techniques
The authors employ averaging and boosting ensemble techniques from machine learning to investigate the predictive power of these features for different types of crimes occurring at the census tract level in New York City. They find that spatial and spatio-temporal features derived from Foursquare venues and check-ins, subway rides, and taxi rides significantly improve baseline models that rely solely on census and point-of-interest (POI) data. The proposed models achieve impressive absolute R^2 metrics, with up to 65% accuracy on a geographical out-of-sample test set and up to 89% accuracy on a temporal out-of-sample test set. This demonstrates that, in addition to the residential population of an area, the ambient population strongly predicts crime levels in that area.
Varying Predictive Power Across Different Types Of Crimes
The authors delve deeper into specific crime categories and observe that the predictive gain of human dynamics features varies across different types of crimes. For instance, these features provide the greatest boost in predicting grand larcenies while assaults are already well predicted by census features alone. Furthermore, Kadar and Pletikosa identify and discuss the top predictive features for each main crime category.
Implications For Urban Policy And Law Enforcement
These findings offer valuable insights for those responsible for urban policy or law enforcement. By leveraging geo-tagged human dynamics data such as public transportation or taxi usage patterns researchers can now empirically measure aspects of criminological theories that were previously difficult to study at a large scale. Overall this research highlights the potential of using large-scale human mobility data for long term crime prediction models . By incorporating these dynamic data sources such as Foursquare venues , check - ins , subway rides ,and taxi rides alongside traditional census data policymakers can enhance their understanding of crime patterns . This will enable them develop more effective strategies to ensure public safety & security .
