Testing the Effectiveness of Geo-Behavioural Profiling Systems

1
Testing the Effectiveness of Geo-Behavioural
Profiling Systems
Professor David Canter Laura Hammond Director,
Centre for Investigative Psychology The
University of Liverpool,UK www.i-psy.com
2
Study of effectiveness of various prioritisation
strategies

• 92 offence series
• in a London borough.
• All identified series
• over four years/

3
Various Models Illustrated from the crimes of
Jack the Ripper
4
The canonical map
Chapman 8.9 6.00
Nichols 31.8 3.40
Kelly 9.11- 7.00?
Stride 30.9 1.00
Eddowes 30.9 1.45
5
The canonical map
6
Circle Hypothesis Calculation to Locate Rippers
Base
7
Central Circle in Whitechapel
8
Centre of Gravity of Murder Locations
9
Two further concepts

• Criminal Range
• Decay Functions

10
Criminal range
11
The Distance Decay Function, An example from
serial arsonists
35
30
25
20
15
10
5
0
0-0.1
0.51-0.99
1.51-2.00
2.51-3
4.01-5
6.01-7
10
0.11-0.5
1.00-1.5
2.01-2.5
3.01-4
5.01-6
7.01-8
Range of Distances traveled by Arsonists
12
Optimal Function of Distance from Offence to
Home
13
Offence Map Developing the Search Tool that
incorporates various distance functions.
1st Crime
2nd
4th
3rd
5th
14
The system assigns a priority to each known
offender using various criteria.
Prioritised search area
Offender ID Address Probability MO Match
124 Location A 0.28574311864 0
427 Location B 0.27038233898 0
427 Location C 0.26035169492 0
226 Location D 0.25577861017 0
48 Location E 0.23282991525 0
124 Location F 0.22445984746 0.3
124 Location G 0.21932662712 0
15
Cumulative Percentages of the Rank Achieved
by Actual Offender in Dragnet Prioritization for
different criteria.

• Criteria
• Distance from centre of circle defined by 2
  furthest crimes
• Distance from centre of gravity (centroid)
• Optimum negative log decay function
• Negative log decay function with buffer

50 of actual offenders are in the top 5
prioritized just by geographical location. iOPS
allows this to be improved by also using M.O.
matching.
16
(N 92)
COMPARISON OF 15 DIFFERENT MODELS
Model Mean Rank (S.D) Median Rank
Exponential Function - High B-value 8.86 (8.446) 4
Logarithmic Function -Low B-value 9.04 (8.402) 4.5
Logarithmic Function -High B-value 8.92 (8.350) 5
Centre of Gravity 9.21 (8.256) 5
Logarithmic Function -Optimal B-value 8.95 (8.349) 5
Logarithmic Function with Buffer Zone- Optimal B-value 9.73 (8.674) 5.5
Exponential Function - Low B-value 10.18 (8.411) 7
Exponential Function - Optimal B-value 10.55 (8.461) 7
Exponential Function with Buffer Zone- Optimal B-value 10.61 (8.888) 7
Last Offence 11.11 (8.416) 9
First Offence 11.96 (8.424) 10..5
Centre of Minimum Distance 12.52 (8.066) 14
Centre of Circle 15.15 (7.536) 21
Straight Linear Function 21.00 ( .000) 21
17
Need for more context specific examination of
crime patterns

• Taking account of
• Land use
• Opportunities for crime
• Guardianship
• Targeting
• Transport routes
• Temporal sequences

18
Offender 34 crimes 1 home location
Offender 1 34 CRIMES 1 HOME LOCATION
19
Offender 34 crimes 1 home location
Offender 2 34 CRIMES 1 HOME LOCATION
20
2 Offenders with different distributions of
crimse in relation to Southside Centre in
Wandsworth
Southside Centre
Offender 34 crimes 1 home location
Offender 34 crimes 1 home location
21
(No Transcript)
22
EXAMPLES FROM BURLGARIES IN WANDSWORTH PNC
data 12 offenders All crime types 1992-present
Offender 1 10 CRIMES 2 HOME LOCATIONS
23
Offender 8 24 CRIMES 1 HOME LOCATION
24
Offender 3 20 CRIMES 3 HOME LOCATIONS
25
Offender 2 21 CRIMES 3 HOME LOCATIONS
26
Offender 7 13 CRIMES 2 HOME LOCATIONS
27
Offender 5 19 CRIMES 3 HOME LOCATIONS
28
Offender 11 48 CRIMES 3 HOME LOCATIONS
29
Offender 12 53 CRIMES 2 HOME LOCATIONS