Trace vs. Time: Entropy Analysis and Event Predictability of Traceless Event Sequencing

Process mining offers powerful techniques to analyze real-world event data, aiming to improve processes. Typically, the data is stored and examined in event logs as traces, where each trace contains the sequence of events pertaining to a specific process case. A case can, e.g., represent the management of a customer request or the sequence of events from ordering to delivering a product to a customer in online retail businesses. While this approach allows to analyze and gain insights from complex event data, it also isolates events that in reality are correlated, potentially concealing important process behavior. In this paper, we motivate and conceptualize the approach to describe the observations generated by the underlying system as a single event sequence that is ordered as being executed. We study and compare how much the event order and trace notion affect the entropy rates of different real-life processes. Further, we investigate how predictable next activities in event sequences are. Our study indicates that ordering the events as executed does not necessarily increase the entropy rates of the process. We discuss these findings and their implications for future research.