Accuracy Assessment of Individually Calibrated Journey-To-Crime Geographic Profiling Models

• Main Author: Pal, Tania
• Other Authors: Michael Leitner
• Format: Others
• Language: en
• Published: LSU 2007
• Subjects: Geography & Anthropology
• Online Access: http://etd.lsu.edu/docs/available/etd-11122007-100403/

Technological advances are fundamental to the development of spatial analysis tools and methodologies available and used within the criminal investigative process. This research focuses on one such methodology for serial crime analysis: Journey-to-Crime (JTC) Geographic Profiling (GP). JTC or the study of the travel behavior between an offenderâs residence to and from the crime scene has been a subject of study within criminology for many years. GP, based on such travel behavior, is a spatial analysis and decision support tool that is used by law enforcement agencies to determine or predict the likely location of a serial offenderâs residence or âhavenâ. The tool uses locations of a connected series of crimes and applies various functional distance measures to them which have been avoided by traditional analytical methodologies. GP models are probability density distributions of crime trips, which help to narrow down the geographical search area or the offense domain for an offender. This research uses 135 serial property crime incidents from Baltimore County, Maryland between 1994 and 1997 for three different crime types - auto theft, larceny and burglary. The objective is to analyze the accuracy of individually (i.e., by crime type and distance decay functions) calibrated JTC GP models by comparing them with the default-valued (available in CrimeStat® 3.1) JTC GP models. The JTC GP accuracy assessment is conducted on the following three measurements: a) Euclidean distance error â the straight-line distance between the actual home location and the predicted home location. b) Top profile area â the area of all cells with a probability score equal to or higher than the probability score assigned to the actual haven. c) Hit score percentage â the ratio of the area searched before the offenderâs residence is found, to the total study area. The smaller the value of the above measures, the better the model predicts. Results indicate that for most cases there are no statistically significant differences between the individually calibrated and default valued JTC GP models. Thus it could be concluded that police department and other investigative agencies using CrimeStat® 3.1 will save resources (personnel, time and financial) if they use the default values for the JTC distance decay functions parameters instead of individually calibrating the data while creating GP models for serial offenders.