Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Threat Prediction Model

Multi-Classification: Determine probability of a soft-target attack in France.
by

Joseph Benyam

on 15 September 2016

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Threat Prediction Model

Global Terrorism Threat Predictor
France

Problem…

Mid-year 2016, Europe remains on high alert for terrorism in the wake of attacks in France, Germany, and Turkey. Despite the growing level of instability in Europe, other major events like the Olympics may become soft targets for potential bad actors.

Hypothesis...

Binary Classification:
Using University of Maryland's Global Terrorism Database of terrorism events between 1970 to 2015, predict the probability of soft target terrorist attack will take place in France.

Multi-Classification: Get the feature importances within the GTD dataset and l
ist the countries that have the highest probability of being a target for terrorism; deliver results in a heat map.
Academia
Aviation

Energy
Media & Entertainment
Faith-based Organizations
Hotel Security
Working Group
Maritime
International
Development /NGO

International Organizations and NGO Security Overseas Seminar
Arlington, VA, 20-21 May
2015 OSAC Faith-Based Organizations Working Group Conference
Plano, TX, 1-2 June
XVII Pan American Games
Toronto, Canada, 10-26 July
OSAC College and University Health, Safety, & Security Seminar
Charleston, WV, 23-24 July
OSAC Latin America Regional Council Meeting
Bogota, Colombia, 21-22 October
OSAC Annual Briefing
Washington, DC, 18-19 November
LARC
PARC
ARC
MENA-RC
Staging the Question
Inferences:
OSAC's Threat Responsibility
Crisis Management

Updating Council content areas on OSAC.gov

Help finding speakers for Council meetings

Visiting Councils on the ground

Benchmarking surveys to assess security concerns

Challeneges with Model Development

Putting local Council members in touch with their OSAC POC in the U.S. for OSAC accounts

Emailing “Country Council Start-up Tool Kit,” including sample by-laws and charter

Producing relevant presentations for discussion at Council meetings

2015
Exploratory Analysis
Modeling the Data

Multi-Classification
Nepal
Earthquake

Region 1: Sub-Saharan Africa
(13,434)
Region 2: East Asia
(786)
Region 3: Eastern Europe
(4,892)
Region 4: Southeast Asia
(10,360)
Region 5: Australasia & Oceania
(246)
Region 6: North America
(3,268)
Region 7: South America
(18,628)
Region 8: Middle East and North Africa
(40,422)
Region 9: Central Asia
(548)
Region 10: South Asia
(37,841)
Region 11: Western Europe
(16,020)
Region 12: Central America & Caribbean
(10,337)

Exploratory Analysis
Number of Country Councils: 42


Average Number of Meetings in 2015: 2


Country Councils
with Private-Sector co-Chairs: 19
Summary
Regional Terrorism Breakdown
Tuning the Predictive Model
Attack Counts
by Country

The below visualization represents a distribution of historical terrorism events sorted by twelve regions.
Iraq 16835
Pakistan 11385
India 8800
Afghanistan 8352
Colombia 6650
Peru 4907
Philippines 4645
Turkey 3093
United Kingdom 3004
Thailand 2713
Spain 2689
Sri Lanka 2630
United States 2287
Yemen 2145
Somalia 2084
France 2082
Nigeria 1992
Lebanon 1944
Chile 1835
West Bank and Gaza Strip 1752
South Africa 1731
Russia 1682
Israel 1669
Ukraine 1393
Libya 1376
Italy 1328
Bangladesh 1264
Egypt 1158
Greece 1119
Algeria 1013
Confusion Matrix
Feature Importance
Baseline: Random Forest on four features yields
86% accuracy
not bad... but not good enough

Optimized: Random Forest using all features and Grid Search
parameters yield
98.7 % accuracy
.

Overfit Test: We ran the model on both the test and training
set to evaluate whether the model is overfit.
Probability of Attack By Number of Fatalities
Soft Target Predictions


Afghanistan 0.929239
Albania 0.750000
Algeria 0.962488
Angola 0.973494
Argentina 0.858647
Armenia 0.666667
Australia 0.900000
Austria 0.806818
Azerbaijan 0.928571
Bahamas 0.000000
Bahrain 0.907895
Bangladesh 0.919304
Belarus 1.000000
Belgium 0.819672
Belize 1.000000
Benin 1.000000
Bhutan 1.000000
Bolivia 0.850000
Bosnia-Herzegovina 0.952381
Botswana 1.000000
Brazil 0.849765
Bulgaria 0.833333
Burkina Faso 0.714286
Burundi 0.969267
Cambodia 0.833333
Cameroon 0.975000
Canada 0.815789
Central African Republic 0.945946
Chad 0.964286
Chile 0.930245
Probability of
Successful Attack
Contact Information
Joseph Benyam
Data Science Student

202.560.1894
josephbenyam@gmail.com

LinkedIn: www.linkedin.com/in/josephbenyam
Full transcript