A Self-Organizing Time Map for time-to-event data

Understanding dynamics in multivariate data before, during and after events, i.e. time-to-event data, is of central importance in a wide range of tasks, such as the path to and afterlife of a failure of a financial institution or country and diagnosis of a disease. The main task of this paper is to provide a solution to exploring dynamics across manifold entities in multivariate data paired with a time-to-event dimension. The Self-Organizing Time Map (SOTM) provides means for visual dynamic clustering by illustrating temporal dynamics on a two-dimension plane. Likewise, the SOTM holds promise for illustrating patterns in time-to-event data by simply interchanging the time dimension for a time-to-event dimension. This provides a new approach to visual analysis of patterns in multivariate data before, during and after events of interest. The time-to-event SOTM is illustrated on toy and real-world data. The real-world case illustrates dynamics in macro-financial data before, during and after modern systemic financial crises.