Terrorism group prediction using feature combination and BiGRU with self-attention mechanism

Area of Science:

• Computational Social Science
• Artificial Intelligence
• National Security

Background:

• Terrorism and extremism pose significant global threats to stability and security.
• Identifying terrorist organizations is crucial for effective counter-terrorism strategies.
• Advancements in AI, machine learning (ML), deep learning (DL), and natural language processing (NLP) offer new tools for security analysis.

Purpose of the Study:

• To develop and evaluate an AI framework for classifying and predicting terrorist organizations.
• To leverage historical terrorism data for improved national and regional security.
• To enhance the accuracy and robustness of identifying entities involved in terrorist activities.

Main Methods:

• A novel framework, Bidirectional Recurrent Units and Self-Attention (BiGRU-SA), was developed.
• Textual features from DistilBERT and correlated features were integrated with Global Terrorism Database (GTD) data.
• The Synthetic Minority Over-sampling Technique with Tomek links (SMOTE-T) was used to address data imbalance.

Main Results:

• The BiGRU-SA framework achieved 98.68% accuracy in classifying 36 terrorist organizations.
• High performance metrics were recorded: 96.06% precision, 96.83% sensitivity, 99.50% specificity, and 97.50% MCC.
• The proposed model outperformed ten other comparative models, including traditional ML and DL algorithms.

Conclusions:

• The BiGRU-SA framework demonstrates superior effectiveness and accuracy in classifying and predicting terrorist organizations.
• This AI-driven approach significantly enhances capabilities for national security and counter-terrorism efforts.
• The study highlights the potential of advanced AI techniques in analyzing complex security challenges using large datasets like the GTD.