• Challenge 2: timestampsEvents nee

• Challenge 2: timestamps
Events need to be ordered per case. In principle, such ordering does not require timestamps. However, when merging data from different sources, one typically needs to depend on timestamps to sort events (in order of occurrence). This may be problematic because of multiple clocks and delayed recording. For example, in an X-ray machine the different components have local clocks and events are often queueing before being recorded. Therefore, there may be significant differences between the actual time an event takes place and its timestamp in the log. As a result the ordering of events is unreliable, e.g., cause and effect may be reversed. In other applications, timestamps may be too coarse. In fact, many information systems only record a date and not a timestamp. For example, most events in a hospital are recorded in the hospital information system based on a patient id and a date, without storing the actual time of the test or visit. As a result, it is impossible to reconstruct the order of events on a given day. One way to address this problem is to assume only a partial ordering of events (i.e., not a total order) and subsequently use dedicated process mining algorithms for this. Another way to (partially) address the problem is to “guess” the order based on domain knowledge or frequent patterns across days.