Non-Stationary Model for Crime Rate Inference Using Modern Urban Data

Area of Science:

• Social Science
• Urban Studies
• Data Science
• Criminology

Background:

• Crime is a significant social issue impacting public safety and socioeconomic status.
• Accurate crime rate inference is crucial for effective policy-making and crime reduction strategies.
• Traditional crime prediction models rely on demographics and geographical factors.

Purpose of the Study:

• To investigate the utility of novel urban big data sources for neighborhood-level crime rate inference.
• To compare the performance of models using traditional features versus those incorporating Point-of-Interest and taxi flow data.
• To address the geospatial non-stationarity in crime-feature correlations.

Main Methods:

• Utilized large-scale Point-of-Interest (POI) data and taxi flow data from Chicago, IL.
• Employed traditional demographic and geographical features as baseline predictors.
• Implemented a geographically weighted negative binomial regression (GWNBR) model to account for spatial heterogeneity.

Main Results:

• Incorporating POI and taxi flow data led to significantly improved crime rate inference performance compared to traditional features.
• The enhanced performance was consistent across multiple years, indicating robustness.
• Feature importance analysis confirmed the significance of the new big data features.
• The GWNBR model demonstrated superior performance over the standard negative binomial model, effectively handling spatial variations.

Conclusions:

• Urban big data, specifically POI and taxi flow, offers valuable new perspectives for understanding and predicting crime.
• The proposed GWNBR model effectively captures the complex spatial relationships between urban dynamics and crime rates.
• This data-driven approach provides a more accurate foundation for developing targeted crime prevention policies.