A hybrid CA/MAS model of residential burglary with AHP and GA-based calibration.

摘要

This dissertation presents an innovative approach to the study of residential burglary. A simulation model is built upon the integration of Cellular Automata (CA) and Multi-Agent Systems (MAS), which utilizes journey-to-crime (JTC), social disorganization (SD), and routine activity (RA) theories to predict locations of residential burglary targets. Offenders are implemented as MAS agents on top of CA automata of targets and places. Each offender has a certain motivation to commit a crime, determined by his/her age, race and gender background. Likewise, each possible location has a particular attractiveness to the offender, such as target desirability and place lack-of-guardianship, which are dependent on neighborhood characteristics, such as median income, race composition, commute time and length of tenure. This model employs two novel calibration methods to derive the simulation weight parameters: automatic Analytic Hierarchy Process (AHP) and hierarchical Genetic Algorithms (GA). Unlike traditional AHP where pairwise comparisons are conducted manually and subjectively, automatic AHP calibrates the pairwise comparison scoring to derive parameter weights for the model using a pseudo-binary search based on empirical data. The hierarchical GA calibration exploits the crime model's hierarchical structure to create a modified GA, which incorporates new composite crossover, composite mutation, and infeasibility repair techniques into the residential burglary weight parameter selection process. The hybrid CA/MAS residential burglary model results showed realistic predictions by the automatic AHP and hierarchical GA calibration methods in terms of RMSE and crime pattern analysis. Also, through the analysis of the calibrated weights, it was found that the JTC theory contributed the most toward producing realistic predictions, followed by RA, then SD. Future work on this project may include parallel processing, additional datasets, and new distance measures and neighborhood representations.