Risk Matrix for Violent Radicalization: A Machine Learning Approach.

Area of Science:

• Computational social science
• Behavioral psychology
• Criminology

Background:

• Previous hypothesis-driven studies offered limited explanations for violent radicalization.
• These studies focused on a restricted set of preselected variables, leading to fragmented understanding.

Purpose of the Study:

• To comprehensively analyze all variables in the "Profiles of Individual Radicalization in the United States" database.
• To identify critical predictors of violent extremism using a machine learning approach without prior assumptions.

Main Methods:

• Utilized a machine learning approach on the "Profiles of Individual Radicalization in the United States" database.
• Analyzed 79 variables to identify those critical for predicting violence.

Main Results:

• Identified 19 critical variables predicting violent extremism with 86.3% accuracy.
• Violent extremists typically had criminal, not radical, backgrounds, were radicalized later in life, followed terrorist groups, sought training, and used social media.
• Non-violent extremists had radical family traditions, higher social strata, and led organizations.

Conclusions:

• Developed a risk matrix based on critical characteristics to predict individual-level violence risk.
• Findings differentiate violent from non-violent extremists and inform risk assessment strategies.