Jorge Muñoz-Gama Universitat Politècnica de Catalunya (Barcelona, Spain) Algorithms for Process Conformance and Process Refinement

Outline 9 Sep 2010Process Conformance and Refinement2  Process Mining, Conformance and Refinement  Process Conformance  Related Work and Motivation  Approach  Implementation and Results  Extensions  Process Refinement  Breaking Concurrencies  Supervisory Control Refinement  Future Work and Conclusions

Process Mining 9 Sep 2010Process Conformance and Refinement3 *

Refinement Process Conformance and Refinement 9 Sep 2010Process Conformance and Refinement4 Petri Net A B D A C D Event Log MDT ETC Precision Metric Conformance (Precision) A D C B A C B D Refined Model Locate the inconsistencies Measure the inconsistencies More accurate model

Conformance Dimensions 9 Sep 2010Process Conformance and Refinement5 FitnessPrecisionGeneralizationStructure How much of the observed behavior is captured by the model Minimal structure which clearly reflect the behavior Models with minimal behavior to represent accurately the log Overly precise models which overfit the log

Outline 9 Sep 2010Process Conformance and Refinement6  Process Mining, Conformance and Refinement  Process Conformance  Related Work and Motivation  Approach  Implementation and Results  Extensions  Process Refinement  Breaking Concurrencies  Supervisory Control Refinement  Future Work and Conclusions

Related Work 9 Sep 2010Process Conformance and Refinement7  Precision in the literature  Most related work Rozinat et al. Information System 33 (2008)  Metric for Precision in Petri Nets  Computation of Follows and Precedes relations (Always, Never, Sometimes) of Model and Log.  Measurement based on discrepancies in Sometimes relations  Model relations require a model space state exploration Coverability Graph

Other Approaches and Motivation 9 Sep 2010Process Conformance and Refinement8  Other approaches as language equivalence or bisimilarity are not suitable for Process Conformance  The complete models behavior is required  Goals and Requirements  Precision Dimension  Petri Nets  Avoid the complete state space exploration  Effort needed to obtain an accurate model  Fine-level precision  Locate the precision inconsistencies

Outline 9 Sep 2010Process Conformance and Refinement9  Process Mining, Conformance and Refinement  Process Conformance  Related Work and Motivation  Approach  Implementation and Results  Extensions  Process Refinement  Breaking Concurrencies  Supervisory Control Refinement  Future Work and Conclusions

General Idea: Escaping Edges 9 Sep 2010Process Conformance and Refinement10 Model Behavior Log Behavior Model Behavior Escaping Edges

Conformance Route Map 9 Sep 2010Process Conformance and Refinement11 Model States Log States Traversal MDT Metric A C B D A B D A C D Petri Net Event Log

Log and Model States 9 Sep 2010Process Conformance and Refinement12  Log  Incorporate state information in the log  (Aalst et al. Software and Systems Modeling, 2009)  Past, Unlimited and Sequence  Model  Markings of the Petri Net

Model States and Mapping 9 Sep 2010Process Conformance and Refinement13  Not all the reachable markings (could be infinite)  Only Markings with a Log State mapped on  Log and Model States Mapping   i.e., reached marking after replay state prefix A E D C B p1 p2p3 A B E p4 s1 p1 s2 p2 s3 p3 s4 p4 p … n p1 p2 p3 p4 p5 Markings not explored

Traversal 9 Sep 2010Process Conformance and Refinement14  Log-guided Traversal of Model Behavior  Allowed Tasks :  i.e., actions enabled in that moment  Reflected Tasks :  i.e., actions really executed (thus, annotated in the log) A E D C B p1 p2p3 p4 A E D C B p1 p2p3 p4 A B E A C E BCDBCD A B E A C E BCBC

Traversal (2) 9 Sep 2010Process Conformance and Refinement15  Escaping Edges :  i.e., enabled actions not executed  Precision discrepancies A E D C B p1 p2p3p4 A B E A C E BCDBCD BCBC D

Precision Metric 9 Sep 2010Process Conformance and Refinement16  Take into account the Escaping Edges  Between 0 (imprecise) and 1 (precise)  More frequent traces, more weight in the metric  Independent of Structural dimension  Globally precision  Localizability A P H Z Q I A H I Z A P Q Z

Minimal Disconformant Traces (MDT) 9 Sep 2010Process Conformance and Refinement17  Localizability of precision inconsistencies  i.e., Minimal traces indicating where the model starts to deviate from the log  Algorithm to compute all MDT using Escaping Edges Refinement Analysis Precision MDT A E A B E C D P Q A D C B Refined Petri Net

Outline 9 Sep 2010Process Conformance and Refinement18  Process Mining, Conformance and Refinement  Process Conformance  Related Work and Motivation  Approach  Implementation and Results  Extensions  Process Refinement  Breaking Concurrencies  Supervisory Control Refinement  Future Work and Conclusions

Implementation 9 Sep 2010Process Conformance and Refinement19  ProM 6 Framework  ETConformance Plug-In

Results 9 Sep 2010Process Conformance and Refinement20

Results (2) 9 Sep 2010Process Conformance and Refinement21

Outline 9 Sep 2010Process Conformance and Refinement22  Process Mining, Conformance and Refinement  Process Conformance  Related Work and Motivation  Approach  Implementation and Results  Extensions  Process Refinement  Breaking Concurrencies  Supervisory Control Refinement  Future Work and Conclusions

Invisible Tasks 9 Sep 2010Process Conformance and Refinement23  Enabled Tasks?  C ?  B and C ? A A A C B p3 p4... A C...  Which Marking?  ?  INDETERMINISM (Transitions associated with no event)

Invisible Tasks (2) 9 Sep 2010Process Conformance and Refinement24  Invisible Coverability Graph  Solutions  Union of Enabled  Lazy Invisibles *  One path only  Shortest Invisible Path * Inv 1 Inv 3 Inv 2 A,B A,D C A,C A A D D B B X X X X C C X X *Rozinat et al. Information System 33 (2008)

Duplicate Tasks 9 Sep 2010Process Conformance and Refinement25  Which Task?  B ?  INDETERMINISM  Solutions  e.g. Look-ahead A B B D C... A B C... (Several Transitions associated with the same event)

Variant: States as Markings 9 Sep 2010Process Conformance and Refinement26  States as Prefix  States as Markings A B C A B C A B C CB p1 p2p3 2 Escaping Edges NO Escaping Edges

Variant: Non fitting models 9 Sep 2010Process Conformance and Refinement27  Symmetric to the Escaping Edges (Ee)  Log Escaping Edges (LEe): The points where the log deviates from the model  Fitness instead of Precision Model Behavior Log Behavior Model Behavior Escaping Edges Log Escaping Behavior

Outline 9 Sep 2010Process Conformance and Refinement28  Process Mining, Conformance and Refinement  Process Conformance  Related Work and Motivation  Approach  Implementation and Results  Extensions  Process Refinement  Breaking Concurrencies  Supervisory Control Refinement  Future Work and Conclusions

Future Work: Refinement 9 Sep 2010Process Conformance and Refinement29  Refinement can be performed by a Domain Expert A C B D Petri Net Refined Petri Net A D C B Event Log A E A B E MDT A E A B E B H J G

 Many causes for precision inconsistencies  Common one is Concurrency  Concurrency in the model allowing several possibilities  But not in the log  Idea is to break the concurrency introducing a new place  We need concurrency relations of the Petri net, the log, and check the results of the new model Breaking Concurrencies 9 Sep 2010Process Conformance and Refinement30

 Concurrency: it exists a reachable marking that enables both transitions, and firing one does not disable the other.  Problematic for large nets  Structural Concurrency  Best effort overapproximation for general Petri Nets  Exact for live and bounded Free Choice systems  Polynomial Algorithm  Kovalyov and Esparza, Proc. Intl. Workshop on Discrete Event Sytems, 1996 Breaking Concurrencies: Petri net 9 Sep 2010Process Conformance and Refinement31 A D C B

 Not concurrencies but the absence of them  Firing Causality Matrix:  Firing Causality: Breaking Concurrencies: Log 9 Sep 2010Process Conformance and Refinement32 A B C D 0

 Break the model concurrency with a place Breaking Concurrencies 9 Sep 2010Process Conformance and Refinement33 A B C D A D C B

Supervisory Control 9 Sep 2010Process Conformance and Refinement34  Supervisory Control in Process Mining  Santos et al. Supervisory Control Service (2010) Supervisor Model MDT Abstraction Refined Model

Conclusions and Future Work 9 Sep 2010Process Conformance and Refinement35  New technique for precision between Petri nets and Log.  Avoids models state space exploration.  MDT, indicating the points where the model starts to deviates from the log.  Approach implemented as plug-in of ProM 6.  Breaking concurrencies to improve the precision.  Supervisory Control for precision refinement.

Thank You 9 Sep 2010Process Conformance and Refinement36 Thank You for Your Attention Papers: A fresh look at Precision in Process Conformance Jorge Muñoz-Gama and Josep Carmona Business Process Management (BPM) 2010