Multidimensional scaling and visualization of patterns in global large-scale accidents

Catastrophic events have been commonly referred to as phase transitions in complex systems (CS). This paper proposes an approach based on unsupervised machine learning to identify phases and phase transitions in the dynamics of CS. The testbed is a dataset of causalities and events associated with global large-scale accidents. Multidimensional time-series are generated from the raw data and are interpreted as the output of a CS. The time-series are normalized and segmented in the time-domain, and the resulting objects are used to characterize the behavior of the dynamical process. The objects are compared through a number of distances and the information by the multidimensional scaling (MDS) technique, respectively. The time is displayed as a parametric variable. The generated portraits have a complex nature, with periods of chaotic-like behavior, and are analyzed in terms of the emerging patterns. The results show that the adoption of MDS is a relevant modeling tool using present day computational resources.