Process discovery Inductive Miner Sander Leemans D Fahland

Process discovery: Inductive Miner Sander Leemans D. Fahland W. M. P van der Aalst

Process discovery system log S. J. J. Leemans process model 2

System behaviour Quality Sound Simple Complete log Recorded log fast Fitting Precise = S. J. J. Leemans 3

Results with incomplete logs not fitting not simple not sound α ILP Flower model not precise. Miner Evolutionary Tree Miner Heuristics not fitting not sound not fast 4

System behaviour Outline {<a, b, c>, <a, c, b>, <a, d, e, f, d, e>} = S. J. J. Leemans 5

Block-structured Petri nets a τ τ b τ x c d τ e / a b → c e d Sander Leemans 6

System behaviour Outline = S. J. J. Leemans 7

Divide & conquer a {<a>, <a>} {<a, b, c>, <a, c, b>, <a, d, e, f, d, e>} recurse S. J. J. Leemans {<b, c>, <c, b>, <d, e, f, d, e>} recurse 8

Finding operator b a {<a, b, c>, <a, c, b>, <a, d, e, f, d, e>} a c d e f {<b, c>, <c, b>, <d, e, f, d, e>} • Find cut in directly-follows graph recurse • Sequence: edges crossing oneway only S. J. J. Leemans 9

recurse … {b, c} a x {<b, c>, <c, b>, <d, e>, {<b, c>, <d, e, f, d, e>} <c, b>} <d, e, f, d, e>} { b c d e f S. J. J. Leemans {d, e, f} , {<b, c> , } <c, b> , <d, e>{, , <d, e, f, d, e>} } Exclusive choice: no crossing edges 10

… one more recursion … a {d, e} x {<b, c>, {<d, e>, <c, b>} <d, e, f, d, e>} {<f>} f d e f S. J. J. Leemans {< < < {f} >, {< d, e >, >, < d, e, f , {< d, e >} >} Loop: identify body and loopback parts (assumption: start/end activities disjoint) 11

… last recursion a {b} x {<b, c>, <c, b>} b c {<b>}{<c>} d {c} {< >{< b, c >, {< > < >}< c , b >}< >} f b e Parallel: all possible crossing edges c S. J. J. Leemans 12

Result b τ c a a x τ τ d e τ f b c d f e S. J. J. Leemans 13

No cut x a x y x z y b z c S. J. J. Leemans 14

Inductive Miner ? • Divide activities • Select operator {a, b} – Else: flower model {c, d} • Split log • Recurse on splitted logs ? {c} Sander Leemans {d} 15

System behaviour Outline = S. J. J. Leemans 16

Rediscoverability System behaviour Complete log = (language equivalent) = (normal form) x • Block-structured with • Startend activities of loop disjoint • No duplicate activities • No silent activities (τ) • Directly-follows graph complete • Noise-free S. J. J. Leemans 17

Incomplete logs System behaviour Sound Simple Complete log Incomplete log polynomial Fitting Most precise (by framework; bring your own operator) S. J. J. Leemans 18

Future Work Generalise – Block-structured x – Startend activities of loop must be disjoint – No duplicate activities – No silent activities (τ) S. J. J. Leemans – Directly-follows graph complete – Noise-free 19

You have been watching System behaviour Sound Simple Complete log Incomplete log polynomial Fitting Most precise (by framework; bring your own operator) S. J. J. Leemans 20