Discovering Unseen Behaviour from Event Logs

Process mining techniques aim to discover insights into the performance of a business process by analysing its event logs. These logs capture historical process executions as sequences of activity occurrences (events). Often, event logs capture only part of the possible process behaviour because the number of executions can be very large, particularly when many activities are executed concurrently. A highly incomplete event log is problematic because process mining techniques use the event log as a starting point. This paper proposes a technique to discover behaviour from an incomplete log. In order to do so, the presented technique builds distributive lattices from the executions captured in the log, which have well-defined notions of completeness and can be used to discover behaviour from few observations. The paper tests the presented approach in a set of real-life event logs and measures the amount of behaviour that can be discovered.