Pubblicazioni

A Case-based Approach to
Business Process Monitoring
S. Montani1, G. Leonardi1
1 Dipartimento di Informatica, University of Piemonte Orientale, Alessandria, Italy
Business Process Management



BP: collection of related activities which produce a
service or a product
BPM: defining, executing, monitoring and optimizing
BPs
BP Optimization: flexibly change and adapt BP:
 Expected situations (new laws, reengineering)
 Unanticipated exceptions (emergencies)
 BP not correctly applied
Business Process Monitoring

Agile Workflow Technology: solution for process
adaptation and overriding:



Modifications of process schema (by process engineers)
Ad-hoc changes of process instances (by end users)
Business Process Monitoring: highlighting and
organizing non-compliances


Allows verification of process conformance
Suggests long-term changes to deal with non-compliances
Case-based
Business Process Monitoring

Analysis based on retrieval of traces similar to the current one



Trace: Sequence of actions occoured in a process instance
Suggestions on how to modify the default process schema:

Analysing most similar examples of changes in retrieved traces

Inspecting clusters obtained on the base of trace similarity
To perform analysis:

Case definition

Similarity measure
Case Structure

Case: a trace of execution of a given process schema
Process schema: general schema definition suggesting what has to be
IDdone
action start
end
actor


1. Trace:
A1 sequence of
1-1-08
10:30
1-1-08 10:35
Robert
actions actually performed by the working team
2. A2
1-1-08 10:30
1-1-08 12:00
Frank
3. A1
2-1-08 12:00
3-1-08 00:30
Louise
Actions
execution
and stored
with additional
4.
A4ordered by
2-1-08
16:00time, 2-1-08
17:03
Carla
information:
5.
A5
2-1-08 23:45
3-1-08 04:55
Joseph
timestamp of the event,
6. the………

the performer or originator of the event (i.e., the person/resource
executing or initiating the action),

data elements recorded with the event (e.g., the dosage of a drug)
Similarity Measure
Trace T1: A1, A2, A1, A4, A5
 Edit Distance: given two traces T1, and T2,
Trace
T2: A1, A1, A3, A5
 ED (T1, T2) = number of operation needed to obtain T1 = T2
To obtain T1 = T2:
 Edit operations on traces:
 Substitute one action with a different one
- Delete
A2 in trace T1
 Insert a new
- Substitute
A4 action
with A3 in trace T1 (or sub. A3 with A4 in T2)

Delete an action
ED(T1, T2) = 2

Each edit operation has cost set to 1
Taxonomy of Actions

All Actions
In our case, the cost of a substitution
is not always 1:

The cost is a value ϵ [0, 1]
Emergency


Hospitalization
The cost is domain dependent
Diagnostic
Therapies
Diagnostic
Therapies
Actions are grouped
in a taxonomy derived
from domain
knowledge
 C.A.T.
Distance
of actions
(A1, A2) proportional
to distance
in the
Evaluation
N.M.R.
Rehabilitation
taxonomy
Taxonomical Distance
All Actions



γ
Let α andEmergency
β be two actions in taxonomy T;
Let γ be the common ancestor of α and β;
γ
Taxonomical
distance between α and β:
Diagnostic
α
Therapies
Hospitalization
Diagnostic
Therapies
β
β
Evaluation
Rehabilitation
 C.A.T.
N1 = n. arcs in
the path between α N.M.R.
and γ;
 N2 = n. arcs in the path between β and γ;
N.M.R)
Evaluation)
= (3 =+γ (1
3)
+/ (3
1)
(1
3 ++2*0)
1+T2*2)
= 1 = 0.33
 dt(C.A.T.,
N3 = n. arcs
between
and
the+/ root
of
N1 = 1;
3; N2 = 1;
3; N3 = 2
0
Action Weight


Also the cost of an insertion or deletion (indirection) is domain
dependent:

Some indirections can cause important effects (high weight value);

other indirections cause minor impact (low weight value).
Let α be an action, its weight wα is defined as:

wα ϵ [0, 1[
if α generates an indirection

wα = 1
otherwhise
Trace Edit Distance

Let P, Q be two traces of actions; let α and β be two
actions ϵ taxonomy T. Trace Edit Distance d(P, Q) is
defined as:

Where: (e1, ... , ek) transforms P into Q, and:


C(ei) = dt(α, β) if ei is the substitution of α with β
C(ei) = 1*wα if ei is the insertion (or deletion) of α in P
(from Q)
Trace Edit Distance: case 1

B and B’ are comparable: dt(B, B’) ≤ τ:


B and B’ considered aligned
Next actions are checked (C and C’)
Trace Edit Distance: case 2


B and B’ are not comparable: dt(B, B’) ≥ τ, and
A and A’ not comparable:


B is substituted with B’
Dt(B, B’) is added to d(P, Q)
Trace Edit Distance: case 3


B and B’ are not comparable: dt(B, B’) ≥ τ, and
A and A’ comparable, and B and C’ comparable:


B aligned with C’; B’ considered as indirection
1*wB’ is added to d(P, Q)
Trace Edit Distance: case 4


B and B’ are not comparable: dt(B, B’) ≥ τ, and
A and A’ comparable, and C and C’ comparable:


B is substituted with B’
Dt(B, B’) is added to d(P, Q)
Application Example:
The Stroke Domain


Stroke: loss of brain function due to ischemia or blood
hemorrage

Medical emergency, potentially leading to permanent damage or to
death

Number 2 cause of death worldwide (soon becoming number 1)
Stroke patient management in 4 phases:

Emergency management;

Hospitalization;

Discharge;

Follow-up.
A “correct” Patient Management

Emergency management:



Symptoms recognition; C.A.T
Hospitalization in neurological ward/stroke unit

Diagnosis: Neurological evaluation + E.C.G. + other diagnostic
investigations (i.e. chest X.R.)

Therapy: antiaggregants; physical rehabilitation
Discharge – Follow-up: no data available
Retrieval Example:
Query Case Description
Atypical situation:

Emergency:


Stroke diagnosed directly after C.A.T., without need of further
investigations  antiaggregant therapy started immediately
Hospitalization:

Patient condition probably unstable, therefore further investigations:
additional tests and a further anticoagulant therapy.

Physical rehabilitation NOT started
diagnosis
Retrieval Example: Indirection:
No physical
and
therapy started
rehabilitation
even in
in the
Anticoagulant
and
Hospitalization
Case Retrieved
retrieved trace – atypical

antiaggregants are
situation captured
comparable: very close
Query case: #103101 in the traceinDB
the taxonomy
since
Indirections:
less
 Trace DB composed by more than 300
traces
very
similar
effect
investigations
have
been performed


K-Nearest Neighbor retrieval using:

K = 20

Trace Similarity Distance
Retrieval time: 1.2 seconds on an Intel Core2 Duo T9400 –
4Gb ram
Query Case
Most Similar Case
Conclusion and
Future Work

CBR-based BP Monitoring:




Cases are traces of execution
Trace Edit Distance
Helps in adapting BP on the basis of deviations in past traces of
the same BP
Clustering to identify the most frequent changes to the BP schema

Implemented (single linkage clustering), but not fully validated

Integration in the Process Mining tool ProM (W. van der Aalst)

Extension of Trace Edit Distance considering temporal delays