Information Technology Reference
In-Depth Information
-
Gradual Drift:
This refers to the scenario as depicted in Fig. 1(c) where
a current process
M
1
is replaced with a new process
M
2
. Unlike the sud-
den drift, here both processes coexist for some time with
M
1
discontinued
gradually. For example, a supply chain organization might introduce a new
delivery process. However, this process is applicable only for orders taken
henceforth. All previous orders still have to follow the older delivery process.
-
Incremental Drift:
This refers to the scenario where a substitution of pro-
cess
M
1
with
M
N
is done via smaller incremental changes as depicted in
Fig. 1(d). This class of drifts is more pronounced in organizations adopting
agile business process management methodology.
M
n
..
:
..
M
2
M
2
M
2
M
3
M
2
M
1
t
M
1
t
M
1
t
M
1
t
(a)
(b)
(c)
(d)
Fig. 1.
Different types of drifts. (a) sudden drift (b) recurring drift (c) gradual drift
and (d) incremental drift. X-axis indicate time and Y-axis indicate process variants.
Shaded rectangles depict process instances.
4 Approaches to Detecting Drifts in Event Logs
We propose approaches to detect potential control-flow changes in a process
manifested as sudden drifts over a period of time by analyzing its event log.
Detecting drifts in data and resource perspectives and in the contexts of gradual,
recurring and incremental drifts is beyond the scope of this paper.
4.1 Causal Footprints
Event logs are characterized by the relationships between activities. Dependen-
cies between activities in an event log can be captured and expressed using
the
follows
(or
precedes
) relationship. For any pair of activities,
a
and
b
∈
Σ
,
one can determine whether they exhibit either
always
,
never
,or
sometimes
fol-
lows/precedes relationship. If
b
follows
a
in all the traces in an event log, then
we say that
b
always follows
a
;if
b
follows
a
only in some subset of the traces
or in none of the traces, then we say that
b
sometimes follows
a
,and
b
never
follows
a
respectively. Consider an event log
L
=
{
acaebfh
,
ahijebd
,
aeghijk
}
containing three traces defined over
Σ
=
{
a
,
b
,
c
,
d
,
e
,
f
,
g
,
h
,
i
,
j
,
k
}
.Thefol-
lowing relations hold in
:
e
always follows
a
,
e
never follows
b
,and
b
sometimes
follows
a
. The variants of
precedes
relation can be defined on similar lines. The
follows/precedes
relationship is rich enough to reveal many control flow changes
in a process. In the next section, we exploit this relationship and define various
features for change detection.
L
