Footnote:
Nach Informationen von SSRN wurde die ursprüngliche Fassung des Dokuments 2012 erstellt
Description:
Process mining is the research area that is concerned with knowledge discovery from event logs and is often situated at the intersection of the fields of data mining and business process management. Although the term entails a collection of a-posteriori analysis methods for extracting knowledge from event logs, most of the attention in the process mining literature has been given to process discovery techniques, focusing specifically on the extraction of control-flow models from event logs. Process discovery (and process mining in general) faces notable difficulties. One difficulty is that process mining is commonly limited to the harder setting of unsupervised learning, since negative information about state transitions that were prevented from taking place (i.e. negative events) is often unavailable in real-life event logs. We propose a method to artificially generate negative events, based on a technique first formulated in the context of the AGNEsMiner process discovery algorithm. In the original version of the algorithm, a configurable completeness assumption is defined by proposing a window size parameter and a negative event injection probability. We present several improvements to make this completeness assumption less strict, in order to prevent the introduction of falsely induced negative events in cases where an event log does not or can not capture all possible behavior, while still allowing for a complete set of negative events, including those which are derived from non-local, history-dependent behavior