Crime Prediction with Historical Crime and Potential Offender Data Using a Spatio-temporal Cokriging Method

Authors: Hongjie YU*, Center of Integrated Geographic Information Analysis,Sun Yat-Sen University, Lin LIU, Department of Geography, University of Cincinnati;Center of GeoInformatics for Public Security,Guangzhou University, Bo YANG, Department of Sociology, University of Central Florida
Topics: Urban Geography
Keywords: Crime prediction; routine activity theory; potential offenders; spatio-temporal Cokriging method
Session Type: Paper
Day: 4/4/2019
Start / End Time: 9:55 AM / 11:35 AM
Room: Washington 4, Marriott, Exhibition Level
Presentation File: No File Uploaded


Crime prediction using machine learning and data fusion algorithms has been gradually becoming a hot research topic. Most of models mainly rely on historical crime data and other factors related to social-economic environment for improving the accuracy of crime prediction, ignoring other contributing factors related to potential offenders, as stipulated routine activity theories. This study takes into account of the possible impact of potential offenders, identified in routine police stop and question operations, and applies a Spatio-temporal Cokriging (ST-Cokriging) method for crime prediction. The historical crime data is used as the primary variable, and the potential offenders as the co-variable. When confirming the correlation between the primary variable and the secondary variable by spearman rank correlation analysis, separate models are implemented for weekly basis, biweekly basis and quad-weekly basis in the XT police district of ZG city, China. Preliminary results show that co-variable of potential offenders showed a significant correlation with the historical crime data under confidence level of 0.01, which explained that inclusion of potential offenders can make a far-reaching impacts on the criminal activities; The accuracies of the ST-Cokriging models with the co-variable are consistently better than those without the co-variable, suggesting that the consideration of pontential offenders improves the perfromace of crime prediction models.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login