Visualization, Feature Selection, Machine Learning: Identifying the Responsible Group for Extreme Acts of Violence

Mahdi Hashemi, Margeret Hall

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

The toll of human casualties and psychological impacts on societies make any study on violent extremism worthwhile, let alone attempting to detect patterns among them. This paper is an effort to predict which violent extremist organization (VEO), among 14 currently active ones throughout the world, is responsible for a violent act based on 14 features, including its human and structural tolls, its target type and value, intelligence, and weapons utilized in the attack. Three main steps in our paper include: 1) the visualization of the violent acts through linear and non-linear dimensionality reduction techniques; 2) sequential forward feature selection based on the generalization accuracy of three machine learning models-decision tree, and linear and nonlinear SVM; and 3) employing multilayer perceptron to predict the VEO based on the selected features of a violent act. Top-ranked selected features were related to the target type and plan and the multilayer perceptron achieved up to 40% test accuracy.

Original languageEnglish (US)
Article number8520859
Pages (from-to)70164-70171
Number of pages8
JournalIEEE Access
Volume6
DOIs
StatePublished - Jan 1 2018

Fingerprint

Multilayer neural networks
Learning systems
Feature extraction
Visualization
Decision trees
Violence

Keywords

  • Multilayer perceptron
  • SVM
  • decision tree
  • feature selection
  • visualization

ASJC Scopus subject areas

  • Computer Science(all)
  • Materials Science(all)
  • Engineering(all)

Cite this

Visualization, Feature Selection, Machine Learning : Identifying the Responsible Group for Extreme Acts of Violence. / Hashemi, Mahdi; Hall, Margeret.

In: IEEE Access, Vol. 6, 8520859, 01.01.2018, p. 70164-70171.

Research output: Contribution to journalArticle

@article{cc1130098d1c48e197b092b7e9d0e1c2,
title = "Visualization, Feature Selection, Machine Learning: Identifying the Responsible Group for Extreme Acts of Violence",
abstract = "The toll of human casualties and psychological impacts on societies make any study on violent extremism worthwhile, let alone attempting to detect patterns among them. This paper is an effort to predict which violent extremist organization (VEO), among 14 currently active ones throughout the world, is responsible for a violent act based on 14 features, including its human and structural tolls, its target type and value, intelligence, and weapons utilized in the attack. Three main steps in our paper include: 1) the visualization of the violent acts through linear and non-linear dimensionality reduction techniques; 2) sequential forward feature selection based on the generalization accuracy of three machine learning models-decision tree, and linear and nonlinear SVM; and 3) employing multilayer perceptron to predict the VEO based on the selected features of a violent act. Top-ranked selected features were related to the target type and plan and the multilayer perceptron achieved up to 40{\%} test accuracy.",
keywords = "Multilayer perceptron, SVM, decision tree, feature selection, visualization",
author = "Mahdi Hashemi and Margeret Hall",
year = "2018",
month = "1",
day = "1",
doi = "10.1109/ACCESS.2018.2879056",
language = "English (US)",
volume = "6",
pages = "70164--70171",
journal = "IEEE Access",
issn = "2169-3536",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - Visualization, Feature Selection, Machine Learning

T2 - Identifying the Responsible Group for Extreme Acts of Violence

AU - Hashemi, Mahdi

AU - Hall, Margeret

PY - 2018/1/1

Y1 - 2018/1/1

N2 - The toll of human casualties and psychological impacts on societies make any study on violent extremism worthwhile, let alone attempting to detect patterns among them. This paper is an effort to predict which violent extremist organization (VEO), among 14 currently active ones throughout the world, is responsible for a violent act based on 14 features, including its human and structural tolls, its target type and value, intelligence, and weapons utilized in the attack. Three main steps in our paper include: 1) the visualization of the violent acts through linear and non-linear dimensionality reduction techniques; 2) sequential forward feature selection based on the generalization accuracy of three machine learning models-decision tree, and linear and nonlinear SVM; and 3) employing multilayer perceptron to predict the VEO based on the selected features of a violent act. Top-ranked selected features were related to the target type and plan and the multilayer perceptron achieved up to 40% test accuracy.

AB - The toll of human casualties and psychological impacts on societies make any study on violent extremism worthwhile, let alone attempting to detect patterns among them. This paper is an effort to predict which violent extremist organization (VEO), among 14 currently active ones throughout the world, is responsible for a violent act based on 14 features, including its human and structural tolls, its target type and value, intelligence, and weapons utilized in the attack. Three main steps in our paper include: 1) the visualization of the violent acts through linear and non-linear dimensionality reduction techniques; 2) sequential forward feature selection based on the generalization accuracy of three machine learning models-decision tree, and linear and nonlinear SVM; and 3) employing multilayer perceptron to predict the VEO based on the selected features of a violent act. Top-ranked selected features were related to the target type and plan and the multilayer perceptron achieved up to 40% test accuracy.

KW - Multilayer perceptron

KW - SVM

KW - decision tree

KW - feature selection

KW - visualization

UR - http://www.scopus.com/inward/record.url?scp=85056181326&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85056181326&partnerID=8YFLogxK

U2 - 10.1109/ACCESS.2018.2879056

DO - 10.1109/ACCESS.2018.2879056

M3 - Article

AN - SCOPUS:85056181326

VL - 6

SP - 70164

EP - 70171

JO - IEEE Access

JF - IEEE Access

SN - 2169-3536

M1 - 8520859

ER -