Phases and Their Transitions Characterizing the Dynamics of Global Terrorism: A Multidimensional Scaling and Visualization Approach

This paper proposes a technique based on unsupervised machine learning to find phases and phase transitions characterizing the dynamics of global terrorism. A dataset of worldwide terrorist incidents, covering the period from 1970 up to 2019 is analyzed. Multidimensional time-series concerning casualties and events are generated from a public domain database and are interpreted as the state of a complex system. The time-series are sliced, and the segments generated are objects that characterize the dynamical process. The objects are compared with each other by means of several distances and classified by means of the multidimensional scaling (MDS) method. The MDS generates loci of objects, where time is displayed as a parametric variable. The obtained portraits are analyzed in terms of the patterns of objects, characterizing the nature of the system dynamics. Complex dynamics are revealed, with periods resembling chaotic behavior, phases and phase transitions. The results demonstrate that the MDS is an effective tool to analyze global terrorism and can be adopted with other complex systems.