Process mining unsupervised

notes of course “Introduction to Data Science” from RWTH-Aachen in semester winter 19/20, Professor van der Aalst, Willibrordus

Event data visualization

This is a visualization of data of stopping renting house. Every point corresponds to an event. Every event has a position based on time that it took place. Every line corresponds to a particular apartment.

One thing to find is vertical data: on a vertical line, same thing happens. Another thing to find is the envelope of the graph is a curve not a line, which means with time growing there are more and more works to do.

Process discovery

Process - case - event

1. Process: event log
2. Case: one trace
3. Event: position in one of traces

Bottom-up

Petri nets: Model allows for any trace

a b c d

1. execute a : p1 and p2
2. for p1 execute b : p3 ( and e is no longer enabled), so, for p2 execute c: p4
3. execute d.

a c b d
a e d

Places are constraints. Hence, process discovery is just finding places.

Top-down

Input: Event log (log is multiset of traces)
Output: Process Tree

parallel: do all things simultaneously
redo: always start with -> and end with ->, like -> or -> f -> or -> f -> f ->

Directly-follows graph based on event log

Sequence cut

Cut a and d (first a and then blabla and then d)
Left b c e f

Loop cut

Cut e and f ( b and c are always start and end)

Sequence cut

cut f and e

Parallel cut

cut b and c

Tooling

• ProM
• pm4py

Difference with pattern mining

• Discovering end-to-end behavior
• More general