Transcript Overview - Spatial Database Group

Mapping and analysis for public safety: An Overview
Motivation

Crime generators and attractors
Identifying events (e.g. Bar closing,
football games) that lead to increased crime.


Predicting crime events
Identifying location and time where a serial
offender would commit his next crime.


Courtsey: www.startribune.com
Predicting the next target of a burglary offender
Identification of patrol routes

Force deployment to mitigate crime hotspots.
http://www.dublincrime.com/blog/wpcontent/MappingOurMeanStreets.jpg
Scientific Domain: Environmental Criminology
Crime pattern theory
Courtsey: http://www.popcenter.org/learning/60steps/index.cfm?stepNum=16
Routine activity theory and Crime Triangle
Courtsey: http://www.popcenter.org/learning/60steps/index.cfm?stepnum=8
 Crime Event: Motivated offender, vulnerable victim (available at an appropriate
location and time), absence of a capable guardian.
 Crime Generators : offenders and targets come together in time place, large
gatherings (e.g. Bars, Football games)
 Crime Attractors : places offering many criminal opportunities and offenders may
relocate to these areas (e.g. drug areas)
What is spatiotemporal data mining ?
Process of discovering interesting, useful and non-trivial patterns from
spatiotemporal data.
Data mining Tasks
Traditional Data Mining
Frequent patterns (e.g.
Associations, Sequential
association, frequent graphs)
Spatiotemporal data mining
(STDM)
ST Frequent patterns (e.g. ST
Co-occurrence, ST
Sequences and Cascading ST
patterns)
Clustering
Hotspot Analysis
Anomaly detection
ST Outliers
Classification/ Regression
ST Classification /
ST (auto) Regression
STDM pattern families
Co-occurrence Patterns
Hotspots
www.sentient.nl/crimeanabody.html
Spatial outliers: sensor (#9) on I-35
Location prediction: nesting sites
Nest locations
Vegetation
durability
Distance to open water
Water depth
Projects : Mapping and Analysis for public safety
 US DoJ/NIJ- Mapping and analysis for Public Safety
 CrimeStat .NET Libaries 1.0 : Modularization of CrimeStat, a tool for the analysis of crime
incidents.
 Performance tuning of Spatial analysis routines in CrimeStat
 CrimeStat 3.2a - 3.3: Addition of new modules for spatial analysis.
 US DOD/ ERDC/ AGC – Cascade models for multi scale
pattern discovery
 Designed new interest measures and formulated pattern
mining algorithms for identifying patterns from large crime
report datasets.
 US DOD –

Spatial network hotspot discovery
New algorithms to discover hotspots along street networks
CrimeStat




A Spatial statistics software to analyze crime incident locations.
It provides modules for spatial statistics, space-time analysis, finding
patterns:
 Hotspot Analysis
 Spatial Modeling
 Crime Travel Demand
Used widely by law enforcement agencies throughout the country.
Popular among Public Health agencies and research groups throughout the
country.
CrimeStat
 Used by law enforcement all over the country (e.g. Redlands Police
Department, Baltimore County)
 File down loads: Fall 2010 65,875 (Source:
http://www.icpsr.umich.edu/CrimeStat/about.html )
 6 Releases since 1999
Our Contributions
•
Crime Stat Libraries 1.0[1]
–
–
•
Set of .NET components distributed by NIJ
Credits: http://www.icpsr.umich.edu/CrimeStat/files/Documentation_for_CrimeStat_Libraries_1.0.pdf
Crime Stat v 3.2-3.3
– Statistical Simulation functions for Spatial Analysis Routines
– Credits: http://www.icpsr.umich.edu/CrimeStat/files/CrimeStat3.3updatenotesPartI.pdf
•
Scalability to Large Datasets
– Self-Join Index[2]
[1]
http://www.spatial.cs.umn.edu/projects/crimestat-pub/beta/
Pradeep Mohan, Shashi Shekhar, Ned Levine, Ronald E. Wilson, Betsy George, Mete Celik, Should SDBMS support the join index ?:
A Case Study from Crimestat. In Proc. of 16th ACM SIGSPATIAL International Conference on
Advances in Geographic Information Systems (ACM GIS 2008), California, USA,2008.
[2]
Real Crime Datasets
Lincoln, NE Dataset
Real Data
 Years 2002- 2007
 > 40 Crime types
 > 200 Sub types
 Average size of each year ~ 40000
Cascading spatio-temporal pattern (CSTP)
Time T1
Time T2 > T1
Time T3>T2
Aggregate(T1,T2,T3)
a
Bar Closing(B)
Assault(A)
Drunk Driving (C)
 Input: Crime reports with location and time.
CSTP: P1
Output: CSTP
C
 Partially ordered subsets of ST event types.
 Located together in space.
 Occur in stages over time.
B
A
Lincoln, NE crime dataset: Case study
 Is bar closing a generator for crime related CSTP ?
Bar locations in Lincoln, NE
Questions
Observation: Crime peaks around bar-closing!
 Is bar closing a crime generator ?
Bar closing
 Are there other generators (e.g.
Saturday Nights )?
Saturday Night
Increase(Larceny,vandalism, assaults)
Increase(Larceny,vandalism, assaults)
K.S Test: Saturday night significantly different than normal day bar closing (P-value = 1.249x10-7 , K =0.41)
Lincoln, NE crime dataset: Case study
{Assault}
{Bar Closing}
{Vandalism}
Spatial
Gen-CPR
Neighborhood
CPI
Max-CPR
1 Mile
0.0386
0.02283
0.0386
2.5 Miles
0.18491
0.04539
0.18491
Temporal Neighbor Size = 1 hr
Dataset Years 2002-2006
Lincoln, NE crime dataset: Case study
Probability of a Bar closing generating a
crime in Lincoln City = 0.038
Crimes considered: Assault and Vandalism
Probability of a Lincoln city downtown
Bar closing generating a crime = 0.0862
Lincoln, NE crime dataset: Case study
Probability of a Vandalism after Bar
closing in Lincoln City = 0.022
Probability of a Vandalism after a
downtown Bar closing = 0.0397
 Only bar closings that also generate assaults
 Downtown subsetting may decrease/ increase
chances.
Lincoln, NE crime dataset: Case study
Probability of an Assault after Bar closing
in Lincoln City = 0.029
Probability of an Assault after a
downtown Bar closing = 0.021
 Only bar closings that also generate Vandalism
 Downtown subsetting may decrease/ increase
chances
Spatial Network Hotspots
Geometric Hotspot
Network Hotspot