predpol presentation to seattle police department
DESCRIPTION
As part of our investigation into predictive policing, Ali Winston and I asked the city of Seattle for records of any analyses or reviews of PredPol's performance there. The Seattle Police Department sent us the following letter on November 6, 2013:Dear Mr. BondGraham,This letter is in response to your public disclosure request dated September 27, 2013 and received by the Seattle Police Public Disclosure Unit on September 27, 2013 for any and all reviews in Seattle's possession as to the efficacy of Predpol's goods and services and performance under the contract1. There has been no review of Predpols contract performance under its current contract with SPD. Therefore there are no responsive records to this part of your request. 2. There has been no analysis or reports on services or products conducted. Therefore there are no responsive records to this part of your request.Seattle Police Department did reach out to Los Angeles Police Department and Santa Cruz for a review of the product and did review the attached PowerPoint upon visiting LA PD.This concludes the Seattle Police Department’s response to your request.TRANSCRIPT
Seattle Police Department
Predictive Policing
Predictive Policing
• No standard definition • Paradigm • Similar to weather forecasting • Not Minority Report
Types of Crime Analysis High
Low (No) Source: Exploring Crime Analysis (2005)
Small group of incidents
Level of Aggregation
One incident
Pattern focused
Larger group of incidents by area and/or crime type
Large to all incidents by city, sector or crime type
Crime Series
Person or incident focused
Crime Hot-spots
Long-term trends
Broad-level focus
Reports (i.e. dashboards)
Examples
Crime type or place-based focus
www.predpol.com
PredPol is …
• software technology that predicts the time and place of future crimes
• high level mathematics combined with experienced
crime analysts and veteran police using their own knowledge and experience
• more than “rear view mirror” policing
that simply maps past crime
www.predpol.com
the science repeat victimization
near-repeats local search
offender behavior mathematical model
2
4B D B Bt
η ω θωδ∂= ∇ − +
∂
deployment to officers in the field
• lists or maps of predictions in 500’ x 500’ boxes
• distributed at start of shift or on-demand • ‘get in the box’ to disrupt crime during patrols
LAPD Deployment
500’
Santa Cruz PD Deployment
making predictive policing practical
tactical ambiguity tactical clarity
‘Get in the Box’
500’
environmental design problem-oriented community policing
intelligence-led broken-windows micro place-based
Disrupt Crime by…
•Detection •Apprehension •Prevention
Crime Detection
Best chance for being at the right place at the right time
Crime Apprehension
Even being near the right place, near the right time is beneficial
One study found that when the first police car arrived
within 4 minutes of a Burglary in Progress call, an arrest was twice as likely as when they arrived at 6 minutes Arrest at the time of burglary is less time-consuming than detecting them after the offenders have left the crime scene
Crime Prevention
High Opportunity/Low Incentive Low Opportunity/High Incentive
“Opportunity makes the thief” Drug Addicts Bored Teens
policing the risk = reducing the opportunity
double the accuracy with PredPol
LAPD Foothill Deployment Nov 2011-April 2012
predictive policing in Santa Cruz, CA
-19%
Santa Cruz burglaries 1st half 2011 vs. 1st half 2012
how do we know it worked in L.A.?
serious property crime all L.A. divisions
Nov ‘11 to May ‘12
-12%
-12%
L.A. Foothill Division
+0.4%
Rest of L.A.
Standardized: October 15 2011 = October 16, 2010
the impact of PredPol
start of LAPD rollout
2010-2011
2011-2012
Day
-12.8%
Predicting Gun Violence with PredPol (Chicago)
Contributions to Prediction
Prediction Accuracy
PreviousHomicideRobbery
Assualt
Weapons Volation
Battery
Components to Gun Homicide
Homicide Robbery Assault Weapons Violation
Battery
Statistical Model
5.5% 13.9% 23.6% 32.1% 24.8%
Chicago 2009-2011 PredPol
hotspotting (3 week)
random
perc
ent h
omic
ides
pre
dict
ed
2x
PredPol predicts gun violence
• is NOT: • a replacement for knowledge, skills & experience • profiling of individuals • probable cause • new hardware or new hires
• IS:
• assigning probabilities of crime in space and time • focal points for effective use of police resources • use of existing crime data
PredPol
SPD Predicted Offenses
• Residential Burglary • Commercial Burglary • Vehicle Prowl • Auto Theft • Robbery • Assualts
Selected Offenses Future Offenses
• Gun Violence • Homicide
SPD Tactical Measurement • Similar to “Koper Curve” deployment • Time in box = 15 – 20 min every 1 to 2 hrs • Concept – Time in box measured by GPS • Black, Red, Amber, and Green color codes for
dosage or treatment of areas • Green box indicates success
Strategic Plan
Comprehensive Data Driven Approach to
Crime Management
Info view
Directed Patrol
Crimeview Incident Mapping
CAU Tactical Crime
Analysis
Data-Driven Objectives
• Create a situational awareness of citywide crime activity, trends, and emerging issues (hard & soft crime)
• Identify effective crime fighting strategies, tactics, and resources to address those issues
• Ensure follow-up and assessment so that desired results are achieved and sustained, and successful strategies and tactics are replicated