Using Deep Learning Neural Networks to Predict Violent vs. Nonviolent Extremist Behaviors

ABSTRACT

Recent analyses of radicalization processes have shown that extremist attitudes and violent behavior may be related in some cases, but are rarely collinear. It therefore benefits analysts of political violence to leverage tools that assist in the distinction of characteristics that might move an individual towards violence (vs. nonviolence) in support of their beliefs. To this end, the current study explores the efficacy of deep learning neural networks for classifying extremists as potentially violent or nonviolent based on dozens of common predictors derived from various perspectives on radicalization. Specifically, this study uses 337 predictors from the Profiles of Individual Radicalization in the U.S. dataset to populate a neural network with two hidden layers composed of four processing nodes. The model correctly predicted whether an individual engaged in violence (or not) in 94.2 percent of cases, on average. Analyses further identified several predictors that were most important in classifying violent and nonviolent cases. These analyses demonstrate neural networks may be effective tools in the study of radicalization and extremism, particularly regarding the disaggregation of salient outcomes.

KEYWORDS:

• Neural networks
• machine learning
• deep learning
• violent extremism
• terrorism
• radicalization

Acknowledgments

The author would like to thank Aarushi Sahejpal for his input and encouragement on this project. The author would also like to thank the two anonymous reviewers whose comments and suggestions improved the paper immensely.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data for this study are available at https://www.start.umd.edu/data-tools/profiles-individual-radicalization-united-states-pirus.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/09546553.2024.2376639

Notes

1. Padraig Cunningham, Matthiew Cord, and Sarah Jane Delaney, “Supervised Learning,” in Machine Learning Techniques for Multimedia, ed. Matthieu Cord and Padraig Cunningham (Boston: Springer, 2008), 21–49.

2. For a synopsis, See Vladimir Nasteski, “An Overview of the Supervised Machine Learning Methods,” Horizons 4 (2017): 51–62, https://doi.org/10.20544/HORIZONS.B.04.1.17.P05.

3. The general linear model (GLM) describes conventional regression models for which there is a continuous dependent variable and continuous or categorical predictor variable(s).

4. E.g., Justin George, “State Failure and Transnational Terrorism: An Empirical Analysis,” Journal of Conflict Resolution 62, no. 3 (2018): 471–95, https://doi.org/10.1177/0022002716660587.

5. E.g., Sara Doering, Garth Davies, and Raymond Corrado, “Reconceptualizing Ideology and Extremism: Toward an Empirically-Based Typology,” Studies in Conflict & Terrorism 46, no. 6 (2023): 1009–23, https://doi.org/10.1080/1057610X.2020.1793452.

6. Feng-Jen Yang, “An Implementation of Naïve Bayes Classifier” (paper presented at the International Conference on Computational Science and Computational Intelligence, Las Vegas, NV, December, 2018).

7. Beakcheol Jang, Myeonghwi Kim, Gaspard Harerimana, and Jong Wook Kim, “Q-Learning Algorithms: A Comprehensive Classification and Applications,” IEEE Access 7 (2019): 133653–67, https://doi.org/10.1109/ACCESS.2019.2941229.

8. Shan Suthaharan, “Support Vector Machine,” in Machine Learning Models and Algorithms for Big Data Classification (Boston: Springer, 2016), 207–35.

9. Fathali M. Moghaddam, “The Staircase to Terrorism: A Psychological Exploration,” American Psychologist 60, no. 2 (2005): 161–9, https://doi.org/10.1037/0003-066X.60.2.161; Mohammed Amin, “The Conveyer Belt Theory of Radicalisation,” https://www.mohammedamin.com/Community_issues/Conveyor-belt-theory.html.

10. Clark McCauley and Sophia Moskalenko, “Understanding Political Radicalization: The Two-Pyramid Model,” American Psychologist 72, no. 3 (2017): 205–16, https://doi.org/10.1037/amp0000062.

11. Tore Bjorgo and Andrew Silke, “Root Causes of Terrorism,” in Routledge Handbook of Terrorism and Counterterrorism, ed. Andrew Silke (Oxon, UK: Routledge, 2018), 57–65.

12. Robert Agnew, “General Strain Theory and Terrorism,” in The Handbook of the Criminology of Terrorism, ed. Gary LaFree and Josh Freilich (Hoboken, NJ: Wiley, 2016), 121–32.

13. Michael Wolfowicz, Yael Litmanovitz, David Weisburd, and Badi Hasisi, “Cognitive and Behavioral Radicalization: A Systematic Review of the Putative Risk and Protective Factors,” Campbell Systematic Reviews 17, no. 3 (2021): e1174, Table 5.

14. Bjorgo and Silke, “Root Causes of Terrorism,” 57.

15. Ann Swidler, “Culture in Action: Symbols and Strategies,” American Sociological Review 51, no. 2 (1986): 273–86, https://doi.org/10.2307/2095521.

16. Alex P. Schmid, Radicalisation, De-Radicalisation, and Counter-Radicalisation: A Conceptual Discussion and Literature Review. Research paper for the International Center for Counterterrorism (The Hague: ICCT, 2013), https://www.icct.nl/sites/default/files/import/publication/ICCT-Schmid-Radicalisation-De-Radicalisation-Counter-Radicalisation-March-2013_2.pdf.

17. Kumar Ramakrishna, “The Role of Ideology in Radicalisation,” in The Routledge Handbook on Radicalisation and Countering Radicalisation, ed. Joel Busher, Leena Malkki, and Sarah Marsden (Oxon, UK: Routledge, 2024), 71–84.

18. Marc Sageman, Leaderless Jihad (Philadelphia: University of Pennsylvania Press, 2008); Stefan Malthaner, “Social Movement Theory and Research on Radicalisation,” in The Routledge Handbook on Radicalisation and Countering Radicalisation, ed. Joel Busher, Leena Malkki, and Sarah Marsden (Oxon, UK: Routledge, 2024), 99–112.

19. Stefan Malthaner and Peter Waldmann, “The Radical Milieu: Conceptualizing the Supportive Social Environment of Terrorist Groups,” Studies in Conflict & Terrorism 37, no. 2 (2014): 979–98, https://doi.org/10.1080/1057610X.2014.962441.

20. Thomas J. Holt, Joshua D. Freilich, Steven M. Chermak, and Gary LaFree, “Examining the Utility of Social Control and Social Learning in the Radicalizaton of Violent and Non-Violent Extremists,” Dynamics of Asymmetric Conflict 11, no. 3 (2018): 125–48, https://doi.org/10.1080/17467586.2018.1470661; Timothy Clancy, Bland Addison, Oleg Pavlov, and Khalid Saeed, “Root Causes of Radicalization: The Terror-Contagion Hypothesis,” System Dynamics Review 40, no. 1 (2024): e1749, https://doi.org/10.1002/sdr.1749.

21. Per-Olof Wikström and Noémie Bouhana, “Analyzing Radicalization and Terrorism: A Situational Action Theory,” in The Handbook of the Criminology of Terrorism, ed. Gary LaFree and Joshua D. Freilich (Hoboken, NJ: Wiley, 2016), 175–86.

22. E.g., Maarten S. O. De Waele, “Youth Involvement in Politically Motivated Violence: Why Do Social Integration, Perceived Legitimacy, and Perceived Discrimination Matter?” International Journal of Conflict and Violence 8, no. 1 (2014): 134–53, https://doi.org/10.4119/ijcv-3050.

23. Kurt Braddock, “The Utility of Narratives for Promoting Radicalization: The Case of the Animal Liberation Front,” Dynamics of Asymmetric Conflict 8, no. 1 (2015): 38–59, https://doi.org/10.1080/17467586.2014.968794; Sarah L. Carthy, Colm B. Doody, Katie Cox, Denis O’Hara, and Kiran Sarma, “Counter-Narratives for the Prevention of Violent Radicalization: A Systematic Review of Targeted Interventions,” Campbell Systematic Reviews 16 (2020): e1106, https://doi.org/10.1002/cl2.1106.

24. For a review, See Özen Odag, Anne Leiser, and Klaus Boehnke, “Reviewing the Role of the Internet in Radicalization Processes,” Journal for Deradicalization 21 (2019). ISSN: 2363-9849.

25. E.g., Caitlin Clemmow, Sandy Schumann, Nadine L. Salman, and Paul Gill, “The Base Rate Study: Developing Base Rates for Risk Factors and Indicators for Engagement in Violent Extremism,” Psychiatry & Behavioral Science 65, no. 3 (2020): 865–81, https://doi.org/10.1111./1556-4029.14282.

26. See Paul Gill and Emily Corner, “There and Back Again: The Study of Mental Disorder and Terrorist Involvement,” American Psychologist 72, no. 3 (2017): 231–41, https://doi.org/10.1037.amp0000090.

27. Sarah Knight, Katie Woodward, and Gary Lancaster, “Non-Violent Versus Violent Actors: An Empirical Study of Different Types of Extremism” (Report for the Defence Science and Technology Laboratory, UK, 2017), 24–6, https://cris.winchester.ac.uk/ws/portalfiles/portal/341019/821821_Lancaster_ViolentVsNon_Violent_original_deposit_with_set_statement.pdf.

28. Bart Schuurman and Sarah L. Carthy, “Understanding (Non)involvement in Terrorist Violence: What Sets Extremists Who Use Terrorist Violence Apart from Those Who Do Not?” Criminology and Public Policy 23, no. 1 (2024): 119–52, Table 4, https://doi.org/10.1111/1745-9133.12626.

29. Michael H. Becker, “When Extremists Become Violent: Examining the Association Between Social Control, Social Learning, and Engagement in Violent Extremism,” Studies in Conflict & Terrorism 44, no. 12 (2021): 1104–24, https://doi.org/10.1080/1057610X.2019.1626093.

30. Paul Gill and John Horgan, “Who Were the Volunteers? The Shifting Sociological and Operational Profile of 1240 Provisional Irish Republican Army Members,” Terrorism and Political Violence 25, no. 3 (2013): 435–56, https://doi.org/10.1080/09546553.2012.664587.

31. John Horgan, “From Profiles to Pathways and Roots to Routes: Perspectives from Psychology on Radicalization into Terrorism,” The Annals of the American Academy of Political and Social Science 618, no. 1 (2008): 80–94.

32. Knight et al., “Non-Violent Versus Violent Actors,” 24.

33. Andrew Silke, “Becoming a Terrorist,” in Terrorists, Victims, and Society: Psychological Perspective on Terrorism and its Consequences, ed. Andrew Silke (New York: Wiley, 2003), 29–52.

34. See Max Taylor and John Horgan, “A Conceptual Framework for Addressing Psychological Process in the Development of the Terrorist,” Terrorism and Political Violence 18, no. 4 (2006): 585–601, Figs. 1, 2, https://doi.org/10.1080/09546550600897413.

35. Schuurman and Carthy, “Understanding (Non)involvement in Terrorist Violence,” 137.

36. Ibid., 138.

37. See Zara Patel, “Counter-Terror Methods in the UK and their Impact on British Muslim Communities,” in Global Crossroads: Rethinking Dominant Orders in Our Contested World, ed. Sahar Taghdisi Rad (London: IJOPEC, 2020), 169–85.

38. See Randy Borum, “Rethinking Radicalization,” Journal of Strategic Security 4, no. 4 (2011): 1–6; Peter R. Neumann, “The Trouble with Radicalisation,” International Affairs 89, no. 4 (2013): 873–93; For related work, See also Mary Beth Altier, Emma Leonard Boyle, and John G. Horgan, “Terrorist Transformations: The Link Between Terrorist Roles and Terrorist Disengagement,” Studies in Conflict & Terrorism 45, no. 9 (2022): 753–77, https://doi.org/10.1080/1057610X.2019.1700038; John Horgan, Neil Shortland, and Suzzette Abbasciano, “Towards a Typology of Terrorism Involvement: A Behavioral Differentiation of Violent Extremist Offenders,” Journal of Threat Assessment and Management 5, no. 2 (2018): 84–102, https://doi.org/10.1037/tam0000102.

39. Gerard Dreyfus, “Neural Networks: An Overview,” in Neural Networks: Methodology and Applications, ed. Gerard Dreyfus (Paris: Springer, 2005), 1–84.

40. The weights of the arcs connected to the input layer are arbitrarily assigned by the model. They are corrected by the network algorithms and activation functions.

41. An activation function may be applied again at the output layer to transform the data in a manner consistent with the analyst’s needs.

42. Fangyu Ding, Quansheng Ge, Dong Jiang, Jingying Fu, and Mengmeng Hao, “Understanding the Dynamics of Terrorism Events with Multiple-Discipline Datasets and Machine Learning,” PLOS ONE 12, no. 6 (2017): e0179057, https://doi.org/10.1371/journal.pone.0179057.

43. START, “Global Terrorism Database,” http://start.umd.edu/gtd (accessed November 1, 2023).

44. Olufemi A. Odeniyi, Mabel E. Adeosun, and Tayo P. Ogundunmade, “Prediction of Terrorist Activities in Nigeria Using Machine Learning Models,” Innovations 71 (2022): 87–96, https://journal-innovations.com/assets/uploads/doc/33c8a-87-96.16092.pdf.

45. M. Irfan Uddin, Nazir Zada, Furqan Aziz, Yousaf Saeed, Asim Zeb, Syed Atif, Ali Shah, Mahmoud Ahmad Al-Khasawneh, and Marwan Mahmoud, “Prediction of Future Terrorist Activities Using Deep Neural Networks,” Complexity 2020 (2020): Article ID 1373087, https://doi.org/10.1155/2020/1373087.

46. Ghada M. A. Soliman and Tarek H. M. Abou-El-Enien, “Terrorism Prediction Using Artificial Neural Network,” Revue d’Intelligence Artificielle 33, no. 2 (2019): 81–8, https://doi.org/10.18280/ria.330201.

47. Qinghao Li, Zonghua Zhang, and Zhen Shen, “Prediction of Terrorist Attacks Based on GA-BP Neural Network,” IOP Conference Series: Materials Science and Engineering 490, no. 6 (2019): 062081, https://doi.org/10.1088/1757-899X490/6/062081.

48. Henry H. Willis, “Guiding Resource Allocations Based on Terrorism Risk,” Risk Analysis 27, no. 3 (2007): 597–606, https://doi.org/10.1111/j.1539-6924.2007.00909.x.

49. Xiaohui Pan and Tao Zhang, “Machine Learning-Based Target Prediction for Terrorist Attacks,” Journal of Physics: Conference Series 2577 (2023): 012007, https://doi.org/10.1088/1742-6596/2577/1/012007.

50. START, “Profiles of Individual Radicalization in the United States Database,” https://www.start.umd.edu/data-tools/profiles-individual-radicalization-united-states-pirus (accessed March 12, 2024).

51. Note that the PIRUS coding scheme allowed individual observations to be associated with up to three ideological sub-categories.

52. Berliana Devianti Putri, Hari Basuki Notobroto, and Arief Wibowo, “Comparison of MICE and Regression Imputation for Handling Missing Data,” Health Notions 2, no. 2 (2018): 183–6.

53. The decision to use the mice package to perform multiple imputation was twofold. First, it is a common imputation technique for datasets used to build predictive models. Second, and more importantly, research on data imputation versus other methods of dealing with missing values has shown that multiple imputation by chained equations tends to produce more accurate predictive models. See, for example, Maritza Mera-Gaona, “Evaluating the Impact of Multivariate Imputation by MICE in Feature Selection,” PLOS One 16, no. 7 (2021): e0254720, https://doi.org/10.1371/journal.pone.0254720.

54. Throughout the manuscript, when the outcome variable (i.e., engagement in violence versus not) is invoked, it will be represented by the variable name, violent, in italics.

55. Information about the h2o analysis package can be found at https://cran.r-project.org/web/packages/h2o/h2o.pdf.

56. The sigmoid activation function is not available in the h2o package, but the hyperbolic tangent activation function operates similarly, albeit with a range from −1 to +1.

57. To be included, the predictor in question needed to appear in the top 20 in at least three of the ten lists produced by the varimp function.

58. Nagelkerke’s R² serves as a coefficient of determination for logit models. It explains variance in violent based on the ten IVs included in the model.

59. Ding et al., “Understanding the Dynamics of Terrorism Events,” 6.

60. Caleb Buffa, Vasit Sagan, Gregory Brunner, and Zachary Phillips, “Predicting Terrorism in Europe with Remote Sensing, Spatial Statistics, and Machine Learning,” International Journal of Geo-Information 11 (2022): 211–23, https://doi.org/10.3390/ijgi11040211.

61. Soliman and Abou-el-Enien, “Terrorism Prediction Using Artificial Neural Network,” 85–6.

62. Odartey Lamptey, Alexander Gegov, Djamila Ouelhadj, Adrian Hopgood, and Serge Da Deppo, “Neural Network Based Identification of Terrorist Groups Using Explainable Artificial Intelligence,” in Proceedings of the IEEE Conference on Artificial Intelligence (Santa Clara, CA, June 5–6, 2023), 191–2, https://doi.org/10.1109/CAI54212.2023.00090.

63. Jolene Scully Gordon, “A Neural Network Approach to the Prediction of Violence” (PhD diss., Oklahoma State University, 1992).

64. Yuan Y. Liu, Min Yang, Malcolm Ramsay, Xiao S. Lie, and Jeremy W. Coid, “A Comparison of Logistic Regression, Classification and Regression Tree, and Neural Network Models in Predicting Violent Re-Offending,” Journal of Quantitative Criminology 27 (2011): 547–73, https://doi.org/10.1007/s10940-011-9137-7.

65. Michael J. Boyle, “Progress and Pitfalls in the Study of Political Violence,” Terrorism and Political Violence 24, no. 4 (2012): 527–43, https://doi.org/10.1080/09546553.2012.700608.

66. Marc Sageman, “The Implication of Terrorism’s Extremely Low Base Rate,” Terrorism and Political Violence 33, no. 2 (2021): 302–11, https://doi.org/10.1080/09546553.2021.1880226.

67. Paul Gill, “Toward a Scientific Approach to Identifying and Understanding Indicators of Radicalization and Terrorist Intent: Eight Key Problems,” Journal of Threat Assessment and Management 2, no. 3–4 (2015): 187–91, https://doi.org/10.1037/tam0000047.

68. Ibid.

69. Seth G. Jones, The Rise of Far-Right Extremism in the United States (CSIS Research Brief, 2018).

70. Paul Gill and Emily Corner, “Lone-Actor Terrorist Target Choice,” Behavioral Sciences & the Law 34, no. 5 (2016): 693–705, https://doi.org/10.1002/bsl.2268.

71. For a discussion related to ideology’s influence on how one engages in violent behavior, See also John Horgan, The Psychology of Terrorism (Oxon, UK: Routledge, 2014).

72. ISIS accounted for 42.8 percent of the Islamist cases in the PIRUS dataset, 80.6 percent of which were violent.

73. Graeme Wood, “What ISIS Really Wants,” The Atlantic, March 2015.

74. See, e.g., Martha Crenshaw, “The Logic of Terrorism: Terrorist Behavior as a Product of Strategic Choice,” in Terrorism in Perspective, ed. Sue Mahan and Pamela L. Griset (Thousand Oaks, CA: SAGE, 2017), 24–33.

Additional information

Notes on contributors

Kurt Braddock

Kurt Braddock is Assistant Professor of Public Communication at American University where he also holds Faculty Affiliations with the Center for Security, Innovation, and New Technology and the Center for Media and Social Impact.