Process Mining in Casehandling Processes

1
Process Mining in Case-handling Processes

• Dr. Minseok Song
• m.s.song_at_tue.nl
• http//is.tm.tue.nl/staff/msong
• 31-08-2007

2
Table of Contents

• Introduction
• Case handling process
• Process mining in case handling process
• Log clustering
• Summary

3
Business Process Management
???? 2005.08.04
4
Process Mining Overview
2) process model
3) organizational model
4) organizational relations
1) basic performance metrics
5) performance characteristics
7) simulation
6) auditing/security
If then
5
Process mining in case handling processes

• Case handling process
• Focuses on cases (e.g. healthcare process,
  software test process)
• Less structured
• Process mining results (i.e. process models,
  social networks) are usually "Spaghetti-like"
  Diagrams
• Shows actual situations in real life
• But, difficult to read and analyze the diagrams

6
Examples
Case handling process
Structural process

• A log from a municipality in the Netherlands
• Invoice handling process
• 570 instances
• 6616 log lines
• 19 activities
• 1 start event, 1 end event
• 111 originators

• A log from a hospital in the Netherlands
• Gynecological oncology process
• 619 instances
• 3574 log lines
• 51 activities
• 21 start event, 24 end event
• 34 originators (departments)

7
Process models from the logs
invoice handling
gynecological process
8
Social networks from the logs
invoice handling
gynecological process
(density 0.021)
(density 0.138)
9
Process mining in case handling processes
process model
social network
10
Life is divided into the horrible and the
miserable. Woody Allen (1935 - )
11
Process log contains too much information!!!
12
Nothing is particularly hard if you divide it
into small jobs. Henry Ford (1863 - 1947)
13
Log clustering
14
Examples
Process Log
Mining result (process model)
15
(No Transcript)
16
(No Transcript)
17
Trace profiles (clustering criteria)

• Activity, e.g.) A,B,C,D vs. A,E,D
• Transition, e.g.) A?B?C, A?C?B
• Originator
• Data type, data value
• Performance
• Number of events, case duration, min transition
  time, max transition time, average transition
  time, median transition time
• ....

18
Profile example
Activity
Originator
Sequence
Performance
...
19
Clustering

• Clustering traces using clustering algorithms
• Clustering techniques in data mining area
• K-Means Clustering
• Quality Threshold Algorithm
• Agglomerative Hierarchical Clustering
• SOM (Self Organizing Map)
• Parametric Clustering

20
K-Means Clustering

• Divide the points into k clusters
• Minimize the total (Euclidian) distance between
  each point and its clusters center.

21
Quality Threshold Algorithm

• Provide a maximum diameter for clusters
• Build a candidate cluster for each point by
  including the closest point

22
Agglomerative Hierarchical Clustering

• Starts with all items in their own clusters.
• Repeatedly merges the two clusters that are the
  closest, based on certain similarity measure.

23
SOM (Self Organizing Map)

• Single layer feedforward network
• The output syntaxes are arranged in low
  dimensional (2D) grid

24
Parametric Clustering

• Consider one dimension scale values
• number of events, case duration, execution time,
  waiting time, sojourn time

60
40
20
0
case duration
1
2
10
4
5
6
7
8
3
9
11
instances
25
Implementation
ProM framework
26
ProM architecture
27
ProM evolution
28
Log clustering plug-in
29
Case study

• A log from a hospital
• 619 instances
• 3574 log lines
• 51 activities
• 21 start event, 24 end event
• 34 originators (departments)

30
Process model from the original log
31
SOM - activity
32
Process models from the clusters
C(1,0) 414
C(1,1) 95
Treatment process
Diagnosis process
33
Social networks from the clusters
C(1,0) 414
C(1,1) 95
Diagnosis process
Treatment process
34
Summary and future works

• Log clustering
• Extending profiles
• Improving visualization
• Evaluating clustering techniques
• Future work
• Explaining clustering results
• Extending Log clustering plug-in with Decision
  Tree

35
Example clustering based on case duration
lt 3 day
lt 6 day
lt 9 day
lt 12 day
gt 12 day
36
Example decision tree for clustering rules
37
References

• Please visit
• http//www.processmining.org
• more than 60 papers available !!

38
Questions?