Building AI and Machine Learning Technologies for Enhancing Transportation Station Area Safety in San Jose, CA

Details

• Creators:
  Ren, Xiangyu ; Yang, Bo ; Zandiatashbar, Ahoura
• Corporate Creators:
  San Jose State University
• Corporate Contributors:
  Research and Education in Promoting Safety (REPS) Tier-1 University Transportation Center (UTC) ; United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology
• Subject/TRT Terms:
  Artificial Intelligence Crimes Involving Transportation Forecasting Machine Learning Passenger Terminals Transit Safety
• Publication/ Report Number:
  REPS-HU-25-02
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States ; San Jose (California)
• Edition:
  Final Report (January 2024 - August 2025)
• Corporate Publisher:
  Research and Education in Promoting Safety (REPS) Tier-1 University Transportation Center (UTC)
• Abstract:
  Accurate crime prediction is crucial for allocating police resources to reduce and prevent crime, particularly in large cities. Previous research has explored the relationship between social and built environments and street crime, but few studies have combined multi-source data using geo-statistical modeling. In this research, we applied a Spatial-Temporal Cokriging algorithm to predict street crime risk in San Jose, California, by integrating historical street crime data with sociological and built environmental variables. We used time-series data from 2,479 police call records in 2019, including assaults and robberies, as the primary variable. Secondary variables included transit density and walkability data from the Environmental Protection Agency (EPA)'s National Walkability Index. We analyzed crime risks across quad-week periods, predicting separately for weekdays and weekends. Our multi-variable crime prediction model compares outcomes using historical data, socio-economic data, and environmental data. The results reveal distinct crime patterns for weekdays and weekends, aiding in crime prediction for rapidly growing metropolitan areas. The study found that the spatial-temporal geostatistical prediction model performed better for weekday crimes, achieving over 70% accuracy, compared to around 60% for weekend crimes. Including transportation density and socio-economic variables significantly improved prediction accuracy, with socioeconomic variables outperforming environmental ones. Interestingly, the impact of these secondary variables varied by day of the week. The study underscores the importance of incorporating auxiliary data to enhance crime prediction and offers a nuanced understanding of the interplay between socio- economic and environmental factors in crime risk.
• Format:
  PDF
• Funding:
  69A3552348323
• Collection(s):
  University Transportation Centers
• Main Document Checksum:
  urn:sha-512:cb701f0995931568d565b21d088b7458e695410b7687a4f2cff73b913085e1b1040ffe99a7f486e52b1d723b9b926365c5df4087b9a5eec395b42edde7e6d32b
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/86328/dot_86328_DS1.pdf
• File Type:
  [PDF - 1.37 MB ]