Process Mining for Time Series Data

Process mining is an established technique to automatically discover a descriptive model of the execution of a process based on event data. Commonly, the event data is recorded by information systems referring to business events and typically tracked at a higher activity abstraction. Disciplines like economics, engineering, life and natural sciences could gain high benefits from process mining in terms of identifying anomalies in the process or supporting predictive analytic in what is being measured. In this way, process mining based analysis give more insights into data than traditional approaches do setting the focus on data correlations. However, these domains mainly rely on sensor data producing time series information, where the event data does not directly relate to high-level business process concepts. This paper suggests an approach for process discovery on "raw" time series data by leveraging clustering to raise the abstraction level of events. As a use-case, we applied our approach on ocean science data where we used raw sensed time-series data from a simulated seasonal coastal upwelling system. In this way, we can give new insights into the data in terms of the identification of anomalies in the process flow aiming to prevent unintended consequences.