Prologa Jan Vanthienen, K.U.Leuven Modeling decision tables PrologaProloga.
EVENT-BASED REAL-TIME DECOMPOSED CONFORMANCE ANALYSIS Seppe vanden Broucke, Jorge Munoz-Gama, Josep...
-
Upload
noah-fleming -
Category
Documents
-
view
218 -
download
0
Transcript of EVENT-BASED REAL-TIME DECOMPOSED CONFORMANCE ANALYSIS Seppe vanden Broucke, Jorge Munoz-Gama, Josep...
EVENT-BASED REAL-TIME
DECOMPOSED CONFORMANCE ANALYSIS
Seppe vanden Broucke, Jorge Munoz-Gama, Josep Carmona, Bart Baesens, and Jan Vanthienen
CoopIS2014
2
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
3
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
4
Process Models, Process Mining and Conformance Checking
Conformance Checking
Process Mining
Process Models
5
• A process is a series of actions taken in order to achieve a particular end.• e.g., construction permit application, patient path in a hospital,
…
• A process model is a description of a process in a certain level of formality.
• Process models are used to gain insight of the processes, simulate forthcoming future, take educated decisions, …
• Several notations for processes: BPMN, EPC, UML, Petri nets, …
Processes and Process Models
6
Petri Net
• Petri nets: formal, wide extended, mathematical foundation, several approaches, …
Transition
PlaceToken
7
Petri Net
• Petri nets: formal, wide extended, mathematical foundation, several approaches, …
8
Process Modeling based on Experts
MODEL REALITY
PROCESS
DOMAINEXPERTS
?
9
Biased Vision
10
• The basic idea of Process Mining is to extract unbiased knowledge from event logs as an exact representation of the reality.
• Conformance Checking is a sub-discipline of Process Mining where a process model is compared with the reality reflected in the log to assess its quality.
Process Mining
11
Conformance Checking
MODEL REALITY
PROCESS
?
LOGS
How good is the model describing
the reality?
12
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
13
• Detect possible conformance anomalies between the model and the reality on the fly, before it is too late.
• Challenges:• Low Computation Time
• Analysis on the regular basis
• Localize Deviation and Understand Causes• Specially in large models
• Event-based• Not complete case necessary to report a mismatch
Event-based Real-time
14
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
15
Addressing the Challenges
• Challenges: • Fast• Event-based• Mismatches localization and understanding.
• Solution:
Decomposition + Event-based Heuristical Replay
Approach
16
17
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
18
Benefits of Decomposition in Conformance
• Decomposition of models has been proven to improve the computation time in conformance checking. • Munoz-Gama et al. 2014, van der Aalst 2013, …
• Meaningful decompositions improve the comprehension of the conformance violations.• Munoz-Gama et al. 2013, …
Single-Entry Single-Exist (SESE) Decomposition
19
Why SESE?
• Single-Entry Single-Exit Components
• Represent subprocesses within the process• Intuitive for conformance diagnosis• Well-studied problem in the literature, with
linear time algorithms.• Hopcroft and Tarjan 1973, Polyvyanyy 2012, …
ExitEntry
20
SESE Decomposition
SESE: set of edges which graph has a Single Entry node and a Single Exit
node
Refined Process Structure Tree (RPST) containing non overlapping SESEs
• Unique• Modular• Linear Time
21
Cut on RPST
• Partitioning over the RPST• Any cut is a partitioning• Algorithm to partitioning by size (k)
• Details in Munoz-Gama et al. 2014.
22
What are the guarantees?
• What are the guarantees in conformance?
• Valid decomposition: only sharing transitions (not places or arcs)
• van der Aalst 2013 proves that:Theorem: If valid decomposition then no
conformance violations are lost in the decomposition process.
23
SESE and Valid Decomposition• SESE decomposition may produce not valid
decompositions• Exit place of one component is the entry place of
another.
• But it can be a valid decomposition if we apply bridging.
24
What are the guarantees?
• Munoz-Gama et al. 2013 proves that:
• Combine SESEs to obtain more understanding subprocesses
Theorem: SESE + bridging is a valid decomposition and therefore, no conformance violations are lost in
the decomposition process.
25
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
Event Dispatching
26
27
Event Dispatching
• Central dispatcher serves each event to its corresponding submodel(s) to be replayed on it.
• Distributed worker threads, each one in charge of one or more submodels in a concurrently way.
28
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
Replay
29
30
Replay vs Alignments
GLOBAL ALIGNMENT-BASED CONFORMANCE
VS
HEURISTIC REPLAY-BASED CONFORMANCE
31
Align-based Conformance
Log Model
C D E
A B B C
Alignment
E
• Conformance mismatch on the Log
• Conformance mismatch on the Model
A B C D EA C D DB C E
…
A B
A C C D DA B B C E
32
Replay-based Conformance
AA
A BA B
A B BA B
CA B B CA B
• Forward replay based on heuristic decisions
33
• Align-based• Optimal from global point of view• High Computational Cost• Trace oriented
• Replay-based• Event oriented• Low Computational Cost• Heuristic and optimality not guaranteed
Align vs Replay Conformance
34
Heuristic Event-based Replay
• Heuristic Event-based replay based on vanden Broucke et al. 2013.• For each submodel and each case id the
state of the submodel is maintained and evolved.• Given a new event we check if it is
enabled in the current state.• Order: transition mapped to activity,
silent transitions, and non-enabled transitions mapped to activity.
35
Heuristic Event-based Replay
• If several candidates, one-step look-ahead.• Sufficient to resolve most ambiguities.
• However, in real-time no knowledge about future events.• Three alternatives:
• Assume determinism• Full replay of the trace each time a new event is
added, reconsidering decisions. • Combination of both: a part of the trace is “frozen”.
• Decomposition limit the effects of the heuristic decisions.
Algorithm
36
• Details in the paper
37
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
Reporting and Visualization
38
39
Reporting and Visualization
• Actions while system is running
• Two types of actions• Logging of statistics
• e.g., be polled regularly by dashboards or persistent data stores
• Triggers fired once certain criteria is met• e.g., error threshold, violation of high-risk task• Actions: warnings, halting running services, …
Proof-of-concept Implementation
40
MonitoringSubprocesses
EventsStreamed
GlobalView
GeneralStatistics
41
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
Bank Transfer: High Level
42
43
Bank Transfer: Petri Net
Scenario 1: Serial Number Check
44
• Pay in cash subprocess requires 3 concurrent checks before to proceed: • CASN (external), CBSN (bank), CIBSN (consortium)
• External check is suddenly skipped (malfunction or attack)
Scenario 2: Preliminary Profiling
45
• First Receiver Pre-Profiling and then Evaluate Pre-Profiling to take a decision of its risk and requirements.
• Evaluation before proper finishing of profiling (malfunction or attack)
Experimental Comparison
46
• Approaches adapted to event-based for comparison• Not designed nor optimized for that (e.g., grouping,
etc)
47
APPROACH
DECOMPOSITION
CONFORMANCE CHECKING
CONCLUSIONS
EVENT-BASED REAL-TIME CONFORMANCE
EXAMPLE CASE AND EXPERIMENTS
EVENT DISPATCHING
REPLAYREPORTING AND VISUALIZATION
48
Conclusions
• Novel process conformance checking to support real-time monitoring of event-based data.• Phases: 1) Decomposition, 2) Event
Dispatching, 3) Replay, and 4) Reporting and Visualization. • Event-based, Fast, and better
comprehension. • Future work: Other decompositions and
real case studies.
EVENT-BASED REAL-TIME
DECOMPOSED CONFORMANCE ANALYSIS
Seppe vanden Broucke, Jorge Munoz-Gama, Josep Carmona, Bart Baesens, and Jan Vanthienen
CoopIS2014