Using translucent activity relationships frequencies to enhance process discovery

Beyel, Ha y H.; an de Aals , Wil M. P.
A icle — Published Ve sion
Using anslucen ac i i y ela ionships equencies o
enhance p ocess disco e y
P ocess Science
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: Beyel, Ha y H.; an de Aals , Wil M. P. (2025) : Using anslucen ac i i y
ela ionships equencies o enhance p ocess disco e y, P ocess Science, ISSN 2948-2178, Sp inge
In e na ional Publishing, Cham, Vol. 2, Iss. 1,
h ps://doi.o g/10.1007/s44311-025-00010-y
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/323691
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h p://c ea i ecommons.o g/licenses/by-nc-nd/4.0/
Open Access
© The Au ho (s) 2025. Open Access This a icle is licensed unde a C ea i e Commons A ibu ion-NonComme cial-NoDe i a i es 4.0 In e -
na ional License, which pe mi s any non-comme cial use, sha ing, dis ibu ion and ep oduc ion in any medium o o ma , as long as you
gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Commons licence, and indica e i you modi ied
he licensed ma e ial. You do no ha e pe mission unde his licence o sha e adap ed ma e ial de i ed om his a icle o pa s o i . The
images o o he hi d pa y ma e ial in his a icle a e included in he a icle’s C ea i e Commons licence, unless indica ed o he wise in a
c edi line o he ma e ial. I ma e ial is no included in he a icle’s C ea i e Commons licence and you in ended use is no pe mi ed by
s a u o y egula ion o exceeds he pe mi ed use, you will need o ob ain pe mission di ec ly om he copy igh holde . To iew a copy o
his licence, isi h p://c ea i ecommons.o g/licenses/by-nc-nd/4.0/.
RESEARCH
Beyeland an de Aals P ocess Science (2025) 2:15
h ps://doi.o g/10.1007/s44311-025-00010-y
P ocess Science
Using anslucen ac i i y ela ionships
equencies oenhance p ocess disco e y
Ha y H. Beyel1* and Wil M. P. an de Aals 1
Abs ac
E en logs a e he p ima y sou ce o in o ma ion in p ocess mining. An e en log
consis s o e en s, each ha ing h ee manda o y a ibu es: a case iden i ie , an ac i -
i y, and a imes amp. T anslucen e en logs add a ou h manda o y a ibu e: enabled
ac i i ies. These a e ac i i ies o which execu ion was possible besides he execu ed
ac i i y. In o ma ion on enabled ac i i ies can be accessed in ask managemen so -
wa e, in asks execu ed wi hin a desk op en i onmen , o by u ilizing domain knowl-
edge. In o ma ion on enabled ac i i ies is aluable o p ocess disco e y and o he
p ocess mining asks. Fo example, u ilizing he in o ma ion when disco e ing p ocess
models om anslucen e en logs esul s in mo e obus and gene alizable p ocess
models. In ecen wo k, anslucen ac i i y ela ionships we e de ined, and hei appli-
ca ion was shown by ex ending he induc i e mining app oach. This wo k ex ends
he de ined ela ionships by also in oducing equencies. As a esul , he Induc i e
Mine —in equen can be ex ended. Hence, i is possible o use hese ela ionships
in he p esence o noise in he eco ded e en logs. We in oduce h ee anslucen
a ian s o he Induc i e Mine —in equen , each using anslucen ac i i y ela ion-
ships a di e en poin s. Mo eo e , we allow o di e en g aphs when all- h oughs
wi hin he Induc i e Mine occu . Using a i icial and en iched eal-li e e en logs, we
show ha conside ing hese ela ionships, mo e da a-speci ic p ocess models ha s ill
ha e desi ed gene aliza ion capabili ies can be disco e ed.
Keywo ds: T anslucen p ocess mining, T anslucen e en log, P ocess disco e y,
Induc i e Mine , Induc i e Mine –in equen
In oduc ion
P ocess mining deals wi h analyzing e en logs, which can be ex ac ed om o gani-
za ions’ da abases ( an de Aals 2016; Dumas e al. 2018). P ocess mining can be
di ided in o h ee a eas: p ocess disco e y (Augus o e al. 2019a), con o mance check-
ing (Ca mona e al. 2018), and p ocess enhancemen (de Leoni 2022). P ocess disco -
e y echniques aim o au oma ically disco e a p ocess model gi en an e en log. Such
a model aims o ep esen he unde lying p ocess comp ehensi ely. Con o mance
checking desc ibes and quan i ies how well a model co esponds o an e en log. P o-
cess enhancemen combines a p ocess model and an e en log o ex end o imp o e
he p o ided model. Examples a e decision mining and p edic ion. An e en log, he
da a sou ce o hese echniques, is a collec ion o e en s, and each e en consis s o
*Co espondence:
[email p o ec ed]
1 Chai o P ocess and Da a
Science, RWTH Aachen
Uni e si y, Aachen, Ge many
Page 2 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
a leas h ee a ibu es: a case iden i ie , an ac i i y, and a imes amp. Thus, e en
logs only cap u e wha happened — no wha could ha e happened. The in o ma ion
on enabled ac i i ies besides he execu ed ac ion is aluable. E en logs ha con ain
in o ma ion on enabled ac i i ies a e called anslucen e en logs.
Two p ominen ways o ecei ing in o ma ion on enabled ac i i ies a e shown in
Beyel and an de Aals (2022) and Beyel e al. (2024). The i s me hod in ol es
injec ing domain knowledge in o an e en log. Fo example, in o ma ion on enabled
ac i i ies can be added by eplaying an e en log on a p ocess model. The second way
is ela ed o use in e ac ions. When moni o ing a use ’s in e ac ions wi hin a desk op
en i onmen , sc eensho s can be aken. In a sc eensho , enabled ac i i ies besides he
use ’s execu ed ac i i y can be iden i ied.
The e is al eady wo k in he a ea o p ocess disco e y ha uses in o ma ion on ena-
bled ac i i ies. A wo-s ep me hod is p esen ed in an de Aals (2019). The me hod
builds on s a e-based- egion echniques and, hus, is no applicable o la ge eal-li e
da ase s. In Beyel and an de Aals (2024a), anslucen ac i i y ela ionships a e
in oduced ha a e embedded in he Induc i e Mine (Leemans e al. 2013a) using
anslucen di ec ly- ollows g aphs de i ed om he ela ionships. To showcase he
alue o echniques ha u ilize in o ma ion on enabled ac i i ies, conside he exam-
ple desk op en i onmen shown in Fig.1. A e choosing a p oposal, a use has o sub-
mi a inancial and echnical e iew. When bo h a e done, he p oposal ge s app o ed
o ejec ed. Also, asking o a e ision is possible. No e ha we assume o his exam-
ple ha he e is an assignmen be ween use s and p oposals, i.e., one p oposal can
only be wo ked on by exac ly one use . An o e iew o he wo k low is shown in
Fig.2. By eco ding hese in e ac ions and aking sc eensho s o each in e ac ion, he
Fig. 1 Example G aphical Use In e ace (GUI) and wo k low o p oposal app o al
Fig. 2 Pe i ne desc ibing he o iginal wo k low
Page 3 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
app oaches in Beyel and an de Aals (2022) and Beyel e al. (2024) can c ea e ans-
lucen e en logs.
An example log based on his ex ac ion is shown in Table1. By conside ing only he
in o ma ion on execu ed ac i i ies, we obse e ha he e a e h ee cases
a,b,c,d,e
,
a,b,c,d, 
, and
a,b,c,d,g,b,c,d,e
. To showcase he enabled ac i i ies in his ep e-
sen a ion, we w i e
a,bc,c,d,e g
,
a,bc,c,d,e g
,
a,bc,c,d,e g,bc,c,d,e g
. Applying
he Induc i e Mine (IM) (Leemans e al. 2013a) on he anslucen e en log esul s in
he Pe i ne shown in Fig.3a.
The wo k low shown in he disco e ed model is pu ely sequen ial, con as ing he eal
p ocess ha allows o concu ency. Howe e , when we conside he in o ma ion on
enabled ac i i ies, we deno e ha ac i i ies b and c a e enabled when appea ing, and c
emains enabled a e b’s execu ion. This indica es ha hese ac i i ies could be pa allel
o each o he . Applying he wo k p esen ed in Beyel and an de Aals (2024a) leads o
disco e ing he Pe i ne depic ed in Fig.3b. As we can deno e, he beha io is he same
as in he o iginal wo k low (see Fig.2).
Howe e , he app oach in Beyel and an de Aals (2024a) su e s om noise
wi hin he ela ionships. Le us assume he ollowing e en log has been ex ac ed:
a,bc,c,d,e g
,
a,bc,c,d,e g
,
a,bc,c,d,e g,bc,c,d,e g
,
a,bc,c,d,e g,e 
. This e en
log has an issue ha can be caused by a w ong ex ac ion om sc eensho s. Fo he
las ace, a e execu ing g, b and c should be enabled, and one o hem should be
execu ed. Howe e , ins ead, e and a e enabled, and is execu ed. When applying he
wo k p esen ed in Beyel and an de Aals (2024a), he esul ing model (see Fig.4a)
allows o much mo e beha io han cap u ed in he log. Using he echniques shown
in his wo k, we disco e a model ha ep esen s he unde lying p ocess be e (see
Fig.4b). As a esul o his sho coming, we ex end (Beyel and an de Aals 2024a)
Table 1 Example anslucen e en log
E en Case Ac i i y Enabled ac i i ies Times amp
e1
1 a {a} 2024 - 10- 20 13:37:37
e2
1 b {b, c} 2024 - 10- 20 13:37:38
e3
1 c {c} 2024 - 10- 20 13:37:39
e4
1 d {d} 2024 - 10- 20 13:37:40
e5
1 e {e, , g} 2024 - 10- 20 13:37:41
e6
2 a {a} 2024 - 10- 20 13:40:37
e7
2 b {b, c} 2024 - 10- 20 13:40:38
e8
2 c {c} 2024 - 10- 20 13:40:39
e9
2 d {d} 2024 - 10- 20 13:40:40
e10
2 {e, , g} 2024 - 10- 20 13:40:41
e11
3 a {a} 2024 - 10- 20 13:45:37
e12
3 b {b, c} 2024 - 10- 20 13:45:38
e13
3 c {c} 2024 - 10- 20 13:45:39
e14
3 d {d} 2024 - 10- 20 13:45:40
e15
3 g {e, , g} 2024 - 10- 20 13:45:41
e16
3 b {b, c} 2024 - 10- 20 13:47:38
e17
3 c {c} 2024 - 10- 20 13:47:39
e18
3 d {d} 2024 - 10- 20 13:47:40
e19
3 e {e, , g} 2024 - 10- 20 13:47:42
Page 4 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
in mul iple ways. Fi s , we in oduce symme y in anslucen ac i i y ela ionships.
Symme y is used o cons uc he g aphs needed o he disco e y algo i hms. Sec-
ond, we in oduce equencies when inco po a ing anslucen ac i i y ela ionships
in he Induc i e Mine (Leemans e al. 2013a). In his p ocess, we embed anslu-
cen exclusi e-choice ela ionships ha we e no used be o e o il e ou noise and
allow o ep esen ing log beha io in mo e de ail. Also, we u ilize equencies as he
Induc i e Mine —in equen (Leemans e al. 2013b) does. Thi d, when all- h oughs
happen, we allow o using he adi ional di ec ly- ollows g aph o ou anslucen
di ec ly- ollows g aph. As we show in ou e alua ion, by using anslucen ac i i y
ela ionships, ewe p ocess a ian s a e needed o disco e a well- ep esen a i e p o-
cess model ha gene alizes well. Fu he mo e, we p esen how exclusi e-choice ela-
ionships change he unde lying g aph necessa y o he Induc i e Mine —in equen .
Fig. 3 P ocess models disco e ed on he aces
a,bc,c,d,e g
,
a,bc,c,d,e g
,
a,bc,c,d,e g,bc,c,d,e g
. The
example shows ha IM o can disco e he unde lying p ocess using less da a
Fig. 4 P ocess Models disco e ed on he aces
a,bc,c,d,e g
,
a,bc,c,d,e g
,
a,bc,c,d,e g,bc,c,d,e g
,
a,bc,c,d,e g,e 
. Unlike he IM o, he IM o can handle in equen ou lie s

Page 5 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
The emainde o he wo k is s uc u ed as ollows. In he Rela ed wo ksec ion,
ela ed wo k is p esen ed. Subsequen ly, in he P elimina iessec ion, p elimina ies o
his wo k a e p esen ed and de ined. We de ine anslucen ac i i y ela ionships in he
T anslucen ac i i y ela ionshipssec ion, ollowed by he anslucen Induc i e Mine —
in equen in he T anslucen Induc i e Mine —in equen sec ion. We e alua e ou
algo i hms in he E alua ionsec ion and discuss ou app oach in he Discussionsec ion.
Finally, we p o ide a conclusion on ou wo k in he Conclusionsec ion.
Rela ed wo k
This sec ion is di ided in o wo pa s. Fi s , we p esen ela ed p ocess-disco e y ech-
niques. Second, we di e in o he cu en s a e o anslucen p ocess mining.
P ocess disco e y
P ocess disco e y ocuses on unco e ing a comp ehensi e p ocess model ha cap u es
he unde lying beha io e lec ed in a gi en e en log ( an de Aals 2016). Nume ous
p ocess-disco e y echniques add ess his challenge, bu all ely on iden i ying ela ion-
ships be ween ac i i ies eco ded in he e en log. A p ominen echnique is he Induc-
i e Mine (IM) (Leemans e al. 2013a). A key concep in IM is he use o di ec ly- ollows
g aphs (DFGs). I an ac i i y ollows an ac i i y, an edge connec s hem. A DFG is pa i-
ioned h ough mul iple cu s o c ea e sublogs. Fo each pa i ion, a DFG is gene a ed,
and he me hod is applied ecu si ely. The IM is a disco e y algo i hm ha p o ides
gua an ees ha no all disco e y echniques ha e, e.g., p o iding a sound wo k low ne
as ou pu o a pe ec i ness w. . . o he log used as inpu . The Induc i e Mine —in e-
quen (IM ) ex ends he IM. In equen beha io can be il e ed ou by keeping ack o
he numbe o occu ences o he di ec ly- ollow ela ionship. In his p ocess, pe ec i -
ness is no longe gua an eed. The e a e also o he a ian s o he IM amewo k, e.g., an
app oxima e a ian ( an De en e al. 2023).
The Spli Mine (Augus o e al. 2019b, 2020) also uses DFGs as inpu . The Spli Mine
ollows six s eps: Fi s , i cons uc s a DFG and iden i ies sel -loops and sho -loops.
Then, concu ency ela ions a e disco e ed om he DFG. Nex , he DFG is il e ed o
balance i ness and p ecision in he inal p ocess model. In he subsequen s eps, spli
and join ga eways a e iden i ied, and inally, OR-joins a e e ined in o XOR o AND
ga eways.
Ano he app oach is egion-based p ocess disco e y. This disco e y echnique builds
on egion heo y (Eh en euch and Rozenbe g 1990a, b). The e a e wo app oaches: lan-
guage-based egion heo y (Mause and Lo enz 2009) and s a e-based egions ( an de
Aals e al. 2010; Co adella e al. 1998; Solé and Ca mona 2011, 2012, 2013). In gene al,
s a e-based egion p ocess-disco e y echniques a e a wo-s ep app oach. Fi s , a ansi-
ion sys em is c ea ed. Second, minimal egions a e ex ac ed om he c ea ed ansi ion
sys em. Each minimal egion co esponds o a place o a Pe i ne . Disco e y echniques
using egion heo y aim o ind a i ing, p ecise p ocess model.
Fo mo e in o ma ion on a ious p ocess-disco e y echniques, we e e o an de
Aals (2016), an de Aals (2022) and Augus o e al. (2022).
Page 6 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
T anslucen p ocess mining
T anslucen p ocess mining deals wi h using in o ma ion on enabled ac i i ies in he di -
e en a eas o p ocess mining. So a , wo ks on c ea ing anslucen e en logs, p ocess
disco e y, and con o mance checking ha e been published.
As desc ibed ea lie , wo app oaches o ecei ing anslucen e en logs a e desc ibed
in Beyel and an de Aals (2022). The i s app oach uses a Pe i ne and an e en log
as inpu . Based on he eplay o he aces, enabled ac i i ies a e iden i ied a each s ep.
This app oach can be used o u ilize domain knowledge o add in o ma ion on enabled
ac i i ies since he model does no ha e o be disco e ed. The second app oach builds on
empla e ma ching (B unelli 2009). An in e ac ion log is c ea ed by eco ding use in e -
ac ions, whe e each en y is linked o a sc eensho showing he s a e o he sc een. I is
checked whe he a labeled empla e is con ained in he sc eensho , and i so, he label
is added as enabled ac i i y. In Beyel e al. (2024), an au oma ic amewo k o ex ac -
ing enabled ac i i ies om sc eensho s is p esen ed. The amewo k uses mo e ad anced
echniques o de ec elemen s o in e es wi hin a sc eensho and labels hem.
The p e iously men ioned p ocess disco e y echnique deals wi h classic e en da a.
E en da a in anslucen e en logs ha e an addi ional a ibu e, enabled ac i i ies,
allowing sophis ica ed echniques. Two echniques ha e been p oposed o p ocess dis-
co e y using anslucen e en logs.
In an de Aals (2019), a s a e-based egion app oach is used. Each se o enabled
ac i i ies in a anslucen e en log ep esen s i s own s a e. A cs connec ing hese s a es
a e labeled wi h he execu ed ac i i y ha ansi ions one se o ano he , as indica ed by
he o de o e en s wi hin a case in he anslucen e en log. In he inal s ep, his sys-
em is ans o med in o a Pe i ne . Howe e , since his app oach elies on s a e-based
egions and hus alls unde egion-based p ocess disco e y, i may encoun e simila
p oblems, such as ex ended compu a ion imes o o e ly complex models.
In Beyel and an de Aals (2024a), he ounda ion o his wo k, successi e e en s
wi hin aces in a anslucen e en log a e used o de i e anslucen ac i i y ela ion-
ships ha ex end di ec ly- ollows g aphs o he IM. The wo k de ined h ee ac i i y ela-
ionships: anslucen di ec ly- ollow ela ionship, anslucen pa allel ela ionship, and
anslucen exclusi e-choice ela ionship. Each ela ionship checks he execu ed ac i i y
and enabled ac i i ies o an e en and he enabled ac i i ies o he subsequen e en . The
execu ed ac i i y has a anslucen di ec ly- ollow ela ionship wi h all subsequen ena-
bled ac i i ies. Also, he execu ed ac i i y has a anslucen pa allel ela ionship wi h all
ac i i ies ha we e enabled du ing and a e he execu ion. Finally, he execu ed ac i i y
has a anslucen exclusi e-choice ela ionship wi h all ac i i ies ha we e enabled du -
ing and a e no enabled a e he execu ion. Also, all ac i i ies a he s a o end o aces
a e conside ed anslucen s a o end ac i i ies. The anslucen di ec ly- ollow and
pa allel ela ionship, as well as anslucen s a o end ac i i ies, a e used o add a cs in
he DFG used by he IM (Leemans e al. 2013a). Th ee a ian s o he IM a e in oduced
in Beyel and an de Aals (2024a) ha use he in o ma ion a a ious poin s. The i s
a ian only uses he DFG en iched wi h in o ma ion on enabled ac i i ies (IM o). The
second a ian i s uses he DFG en iched wi h in o ma ion on enabled ac i i ies, and
i no cu is de ec ed, a classic DFG is used (IM ). The hi d a ian i s uses he classic
DFG, and i no cu is de ec ed, a DFG en iched wi h in o ma ion on enabled ac i i ies
Page 7 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
is used (IM s). As shown in Beyel and an de Aals (2024a), when only a ac ion o he
da a is a ailable, inco po a ing hese ela ionships in he disco e y p ocess helps o ind
mo e obus p ocess models. This is o high in e es when conside ing ha only a ac-
ion o da a om a sys em is used o disco e he p ocesses con ained in he da a.
In his wo k, we ex end he app oach p esen ed in Beyel and an de Aals (2024a)
in se e al ways. Fi s , we de ine symme ic ela ionships. These ela ionships allow o
cap u ing a holis ic iew o ac i i y ela ionships. Second, we in oduce equencies o
allow o mo e obus disco e y. Mo eo e , by u ilizing equencies o he occu ence
o hese ela ionships in he da a, we can u ilize he anslucen exclusi e-choice ela-
ionship, which has been unused be o e. We use hem o il e ou noise in anslu-
cen di ec ly- ollows and pa allel ela ionships. Consequen ly, his wo k unlocked a
g ea e po en ial o using anslucen ac i i y ela ionships han he p e ious wo k.
In addi ion, when u ilizing equencies, we allow o il e ing in equen ela ionships
be ween ac i i ies, simila o he Induc i e Mine —in equen (IM ) (Leemans e al.
2013b). Thi d, we allow use s o choose a classic DFG o a DFG en iched wi h in o -
ma ion on enabled ac i i ies when applying all- h oughs.
In Beyel and an de Aals (2024 d), a p ecision me ic o Pe i ne s and ans-
lucen e en logs is p esen ed. The me ic is based on escaping a cs (Munoz-Gama
and Ca mona 2010). Pa allelism in a model lowe s he p ecision sco e by in oducing
mo e escaping a cs i no co e ed in classical e en logs. Thus, he men ioned wo k
in ol es enabled ac i i ies in he compu a ion o allow o some escaping a cs i hey
a e p esen in he anslucen e en log.
P elimina ies
In his sec ion, we de ine he p elimina ies o ou wo k. We s a wi h basic no a ions.
De ini ion 1 (Basic No a ions). Gi en a se X,
B(X)
deno es he se o all mul i-
se s o e se X. E.g., i
X={x,y,z}
, a possible bag is
[x,x,y]=[x2,y]
. The conjunc-
ion o wo mul ise s is deno ed wi h
⊎
, e.g.,
[x2,y
]⊎[
x,z
]=[
x3,y,z]
. Gi en a mul ise
M∈B(X)
, |M| deno es he size o he mul ise . We de ine
|[ ]| = 0
, and
|A⊎[x]|=|A|+1
.
Fo example,
[x3,y,z]=5
. We deno e he Ca esian p oduc o wo se s X and Y as
X×Y={(x,y)|x∈X∧y∈Y}
. Gi en a se X, a sequence
σ∈X∗
o leng h
n∈N
assigns an enume a ion o elemen s o he se , i.e.,
σ:{1, ..., n}→X
. We deno e his wi h
σ=�σ1,...,σn�
, i.e.,
σi
o he sequence’s i- h elemen .

is he emp y sequence. Gi en a
sequence
σ
,
|σ|
deno es he leng h o a sequence. Gi en wo sequences
σ=�σ1,...,σ|σ|�
and
σ′
=�σ
′
1,
...
,
σ
′
|σ
′
|�
, a conjunc ion is deno ed as
σ
·σ
′
=�σ1
,
...
,
σ|σ|
,
σ
′
1,
...
,
σ
′
|σ
′
|�
. A
sequence
σ′
is a subsequence o sequence
σ
, w i en
σ′⊑σ
, i
σ′=�σl,σl+1,...,σm�
and
1≤l<m≤|σ|
. Gi en
σ∈X∗
, we de ine
�σ0,...,σ0�=��
, and
�σ0,...,σn�=�σ1,...,σn�
.
We conside anslucen e en logs as inpu o ou p ocess-disco e y echniques,
i.e., o each e en , in o ma ion on enabled ac i i ies besides he execu ed ac i i y is
a ailable. Impo an o no e is ha he execu ed ac i i y o an e en is also an enabled
ac i i y in he co esponding e en .
Page 8 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
De ini ion 2 (T anslucen E en Log).
Uac
is he uni e se o ac i i y names. An e en
e is a uple
e=(en,ac )∈P(Uac )×Uac
such ha
ac ∈en
. We de ine
πen(e)=en
and
πac (e)=ac
. A ace
σ=�e1,...,en�∈(P(Uac )×Uac )∗
is a sequence o e en s. A
anslucen e en log
L∈B((P(Uac )×Uac )∗)
is a mul ise o aces. Fo simplici y, we
de ine
πac (L)=σ∈Le∈σ{πac (e)}
. Mo eo e , we assume
σ∈Le∈σπen(e)=πac (L)
.
Fu he mo e, |L| deno es he numbe o aces in L.
Le
Lex =[σ1,σ2,σ3] = [�({a},a),({b,c},b),({c},c),...�,...]
be an example ans-
lucen e en log. Then,
πac (({b,c},b)) =b
and
πen(({b,c},b)) ={b,c}
. In addi ion,
πac (Lex)={a,b,c,...}
. In he emainde o his wo k, we use a sho hand no ion o
which lis all ac i i ies and unde line he execu ed ac i i y:
Lex = [�a,bc,c,...�,...]
.
Also,
|Lex|=3
.
Fu he mo e, we de ine he p ojec ion o anslucen aces. In addi ion o he clas-
sical p ojec ion o execu ed ac i i ies, he enabled ac i i ies a e also p ojec ed. S ill,
an e en is no p ojec ed i he execu ed ac i i y is no pa o he gi en se o ac i i-
ies. The ollowing de ines ha .
De ini ion 3 (T anslucen T ace P ojec ion). Le
L∈B((P(Uac )×Uac )∗)
be a ans-
lucen e en log and
�(X1,x1),(X2,x2),...,(X|σ|,x|σ|)�=σ∈L
be a anslucen ace. Le
A⊆πac (L)
be a se o ac i i ies.
σ↾A
is he anslucen p ojec ion o
σ
on o a se A, i.e.,
σ↾A=�(Xi∩A,xi)|(Xi,xi)∈σ∧xi∈A�
.
Fo example,
�({a},a),({b,c},b),({c},c)�↾{a,b}=�({a},a),({b},b)�
.
Ou wo k ex ends he IM (Leemans e  al. 2013b) o conside in o ma ion on
enabled ac i i ies. To allow o he inco po a ion o his in o ma ion, we in oduce
di ec ly- ollows g aphs, anslucen di ec ly- ollows g aphs om Beyel and an de
Aals (2024a), and sho ly explain he IM .
A di ec ly- ollows g aph is de i ed om a ( anslucen ) e en log and shows h ee
aspec s: s a ac i i ies, end ac i i ies, and di ec ly- ollows ela ionships be ween
ac i i ies eco ded in he p o ided e en log. Di ec ly- ollows ela ionships show
which ac i i ies di ec ly- ollows an ac i i y o e all aces in an e en log. The in o -
ma ion is collec ed in a g aph whe e nodes ep esen s a and end and he ac i i ies
in a log. Edges connec he s a node and ac i i y nodes i aces s a wi h hem.
Edges connec ac i i y nodes wi h he end node i aces and wi h hem. Two ac i i-
ies a e connec ed i one ac i i y is ollowed by ano he ac i i y.
De ini ion 4 (Di ec ly-Follows G aph). Le
L
∈
B((P(Uac )
×
Uac )∗)
be a anslu-
cen e en log wi h
/
∈
πac (L)
and
◮/
∈
πac (L)
. A Di ec ly-Follows G aph (DFG) o L
is a di ec ed g aph
DFGL
=
(
V,E
)
o which
V=πac (L)∪{◮,�}
, whe e
◮
is he s a
node,

is he end node, and
E
⊆({◮}×
σ∈L
{πac (σ1)})∪(
σ∈L
{πac (σ|σ|)}×{�}
)
∪({◮}×{�})

∪
(
σ∈
L

1
≤
i
<|σ|
{(πac (σi),πac (σi+1))}
)
. No e ha he emp y ace (

) can
be cap u ed wi hin his g aph.
The DFG o he anslucen e en log
L= [�a,b
c,c,d,e g�
,
a,b
c,c,d,e g
,
�a,b
c,c,d,e g,bc,c,d,e g�]
is p o ided in Fig.5a. As deno ed in he DFG, ac i i y a is
s a ac i i ies, and e and a e end ac i i ies. Fu he mo e, a is connec ed o b, and b
is connec ed o c, and c o d. d is connec ed o e, , and g. g is connec ed o b.
Page 15 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
in o ma ion on enabled ac i i ies. The second app oach (IM ), showcased in Fig.8b,
s a s wi h a base case. I he e is no base case, a DFG is c ea ed. I no cu is ound, a
DFG is c ea ed. I no cu is de ec ed, a classic DFG is c ea ed. I again no cu is ound,
he DFG ge s il e ed by equency as de ined by he IM (Leemans e al. 2013b). The
hi d app oach (IM s) is simila o he second, bu in o ma ion on enabled ac i i ies is
conside ed a e he classical app oach ails. In he case o all- h oughs, we allow o
he usage o anslucen and classic DFGs. In he ollowing, we show how o inco po-
a e equencies and how he spli s in he equency se ing a e de ined.
T anslucen equen DFG
As he DFG, in oduced in Beyel and an de Aals (2024a) and p esen ed ea lie , he
anslucen equen DFG ( DFG) u ilizes in o ma ion om anslucen ela ionships.
While he anslucen exclusi e-choice ela ionship is unused in he DFG, i is used in
his app oach o lowe connec ion equencies. This u he il e s ou noise. In he ol-
lowing, we de ine he se o a cs ha a e la e used in ou DFG.
Fi s , we de ine equen anslucen di ec ly- ollow a cs. I is coun ed as how o en a
anslucen di ec ly- ollow ela ionship be ween wo ac i i ies occu s in a gi en ans-
lucen e en log. The coun is sub ac ed by he numbe o imes an exclusi e-choice
ela ionship was obse ed be ween hese wo ac i i ies o il e ou noise. As o he IM
(Leemans e al. 2013b), a use can de ine a h eshold . Only ela ionships be ween wo
ac i i ies a e kep i hey occu equen ly enough.
De ini ion 14 (F equen T anslucen Di ec ly-Follow A cs). Le
L∈B((P(Uac )×Uac )∗)
be a anslucen e en log, and
a,b∈πac (L)
. We de ine
→
L
eq
(a,b)=d
L
eq(a,b)−excL
eq_synch
(a,b
)
. Le
0≤ ≤1
. The se o equen anslu-
cen di ec ly- ollows a cs is de ined as
EL,
d
={(a,b)∈πac (L)×πac (L
)
|
→L
eq
(a,b)>0∧→
L
eq
(a,b
)>
 ·maxc∈πac (L)→
L
eq
(a,c)
}
.
Le
L= [�a,bc,c,d,e g�
,
a,bc,c,d,e g
,
�a,bc,c,d,e g,bc,c,d,e g�,�a,bc,c,d,e g,e �]
be again he p e ious example anslucen e en log and
=0
. Then we deno e
→L
eq
(a,b)=
4
,
→L
eq
(b,c)=
5
,
→L
eq
(c,b)=
0
, and
→L
eq
(g,e)=1−1=
0
.
Second, we de ine equen anslucen pa allel a cs. As wi h he anslucen di ec ly-
ollows ela ionships, we coun again how o en a pa allel ela ionship occu s. To coun
his, we used he p e iously in oduced symme ic coun . Again, we sub ac by he
numbe o obse ed exclusi e-choice ela ionships o il e ou noise. As o he IM
(Leemans e al. 2013b), and as p e iously men ioned, a use can de ine a h eshold o
which beha io is il e ed ou .
De ini ion 15 (F equen T anslucen Pa allel A cs). Le
L∈B((P(Uac )×Uac )∗)
be a
anslucen e en log, and
a,b∈πac (L)
. We de ine
+L
eq
(a,b)=pa
L
eq
_
sym
(a,b)
−
excL
eq_synch
(a,b
)
. Le
0≤ ≤1
. The se o equen anslucen pa allel a cs is de ined as
EL,
pa
={(a,b)∈π
ac
(L)×π
ac
(L)
|

+L
eq
(a,b)>0∧+
L
eq
(a,b
)>
 ·maxc∈πac (L)+
L
eq
(a,c)
}
.

Page 16 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
Le
L
= [�
a,bc,c,d,e g�
,
a,bc,c,d,e g
,
a,bc,c,d,e g,bc,c,d,e g
,
�a,bc,c,d,e g,e �]
be again he p e ious example anslucen e en log and
=0
. Then we deno e
+L
eq
(a,b)=
0
,
+L
eq
(b,c)=
5
,
+L
eq
(c,b)=
5
, and
+L
eq
(g,e)=1−1=
0
.
Thi d, we de ine equen anslucen s a a cs. F equen anslucen s a a cs a e
de ined simila ly o he IM (Leemans e al. 2013b).
De ini ion 16 (F equen T anslucen S a A cs). Le
L∈B((P(Uac )×Uac )∗)
be a
anslucen e en log,
◮/
∈
πac (L)
, and
0
≤
≤
1
. The se o equen anslucen s a a cs is
de ined as
EL,
S a
={(◮,a)∈{◮}×π
ac
(L
)
|
S a eq
(a)
L
> ·max
c∈πac (L)
S a
eq
(c)
L}
.
Finally, we de ine equen anslucen end a cs. Also, equen anslucen end a cs
a e de ined simila ly o he IM (Leemans e al. 2013b).
De ini ion 17 (F equen T anslucen End A cs). Le
L∈B((P(Uac )×Uac )∗)
be a
anslucen e en log,
/∈πac (L)
, and
0≤ ≤1
. The se o equen anslucen end a cs
is de ined as E
L,
End
={(a,)∈π
ac
(L)×{
}
| End
eq
(a)
L
> ·max
c
∈π
ac
(
L
)End
eq
(c)
L}
.
By u ilizing he se s o equen a cs, we c ea e a DFG.
De ini ion 18 (T anslucen F equen Di ec ly-Follows G aph). Le
L∈B((P(Uac )×Uac )∗)
be a anslucen e en log,
◮,�/∈πac (L)
, and
0≤ ≤1
. Fu -
he mo e, le
EL,
d
, E
L,
pa
,
EL,
S a
, and
EL,
End
be as de ined be o e. A anslucen equen
Di ec ly-Follows G aph ( DFG) o L and is a di ec ed g aph
G=(V,E)
whe e
V=πac (L)∪{◮,�}
, whe e
◮
is he s a node,

is he end node, and
E
⊆(E
L,
S a
∪E
L,
End
∪E
L,
d
∪E
L,
pa ∪{(◮,�)}
)
.
To illus a e he di e ence be ween he classic DFG, he DFG om Beyel and an
de Aals (2024a) and in oduced ea lie , and he newly in oduced DFG, conside
Fig.9. While he DFG su e s om only sequen ial beha io , he DFG shows a lo o
pa allel beha io . The DFG il e s he in equen ela ionships ou due o he ans-
lucen exclusi e-choice ela ionships and e lec s he beha io be e . The emo al
allows o he disco e y o be e p ocess models.
Fig. 9 Va ious g aphs based on
L= [�a,bc,c,d,e g�
,
a,bc,c,d,e g
,

a,bc,c,d,e g,bc,c,d,e g

,
�
a,bc,c,d,e g,e
�]
Page 17 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
Log spli ing in il e ed se ing
Spli s on he il e ed DFG a e only de ined o adi ional e en logs. Thus, we ha e
o ex end he spli ing de ini ions. No e ha he concu ency spli s ays he same as
in he anslucen IM (Beyel and an de Aals 2024a). Hence, we only ede ine he
exclusi e-choice, sequence, and edo-loop spli . These a e simila o he spli s in IM
(Leemans e al. 2013b).
To spli a log acco ding o an exclusi e-choice il e ed spli , each ace is pu in o
a sublog. All e en s ha a e no om he ac i i y pa i ion o a sublog a e de ia ing.
The spli aims o maximize he numbe o e en s in a ace ha i s he ac i i y pa i-
ion. No e ha he enabled ac i i ies a e p ojec ed on he pa i ion as well.
De ini ion 19 (Exclusi e-Choice Spli Fil e ed). Le
L∈B((P(Uac )×Uac )∗)
be a anslucen e en log and
(×,A1,...,An)
be an exclusi e-choice cu .
Then, he exclusi e-choice spli il e ed is o
i∈{1, ...,n}
de ined as
L
i
={σ↾Ai∈L|i=a g max
j∈{1,...,n}
|π
ac
(σ )↾Aj|}
1.
Fo example, he log
L= [�ac,bc�,�c,c,c�,�ac,bc,c�]
and he cu
(×,{a,b},{c})
would
esul in
L1= [�a,b�2]
and
L2= [�c,c,c�]
.
The sequence spli il e ed minimizes he numbe o e en s ha a e de ia ing. To do
so, a unc ion indSpli is in oduced. The unc ion akes he ace, he ac i i y pa i ion
o in e es , an index indica ing he posi ing wi hin he ace, and he p e ious ac i i y
pa i ions, which should be igno ed, as inpu . The unc ion e u ns he index o which a
spli poin is iden i ied. Each iden i ied segmen is hen p ojec ed on he assigned ac i -
i y pa i ion.
De ini ion 20 (Sequence Spli Fil e ed). Le
L∈B((P(Uac )×Uac )∗)
be a anslu-
cen e en log and
(→,A1,...,An)
be a sequence cu . We de ine
indSpli (σ ,A,s,I)=
min a g min1≤j≤|
σ
|
|
{e∈�σs,...,σj�|πac (e)/∈I}
|
−|{e∈�σs,...,σj�|πac (e)∈A}|
.
We de ine
s1=0
. Also, we de ine
ei
=
indSpli
(σ
,Ai,si,I)
. Mo eo e , we de ine o
i∈{2, ...,n}
:
Fu he mo e, we de ine o each
σ∈L
,
s,e∈{0, ...,|σ|},s≤e:
Then, he sequence spli il e ed is o
i∈{1, ...,n}
de ined as
Li={ge SubSequence(σ ,si,ei)↾Ai|σ∈L}
.
Gi en he log
L= [�ac,bc,c�,�bc,ac,c�,�c,ac,bc,c�]
and he cu
(→,{a,b},{c})
, he
esul is
L1= [�a,b�2,�b,a�]
and
L2= [�c�3]
. The i s c o he hi d ace o L is emo ed.
Also, again, a p ojec ion o enabled ac i i ies akes place.
s
i=

ei−1, i si−1=ei−
1
ei−1+1, i si−1�= ei−
1
ge SubSequence(σ ,s,e)=

��, i s=
e
�σs,...,σe�, i s�=
e
1
a g max
e u ns one
j∈{1, ...,n}
e en when when he a e mul iple alues ha maximize
|π
ac
(σ )↾
A
j|
Page 18 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
The edo-loop spli il e ed de ec s de ia ions when a ace ends wi h an ac i i y no
pa o he i s pa i ion. I his occu s, an emp y ace is added o he log o he i s
pa i ion.
De ini ion 21 (Redo-Loop Spli Fil e ed). Le
L∈B((P(Uac )×Uac )∗)
be a anslu-
cen e en log and
(,A1,...,An)
be a edo-loop cu . Fo
σ∈L
we de ine
Mo eo e , o
σ∈L
we de ine
The edo-loop spli is o
i∈{1, ...,n}
de ined as
L
i={σ↾A
i
|σ
′·
σ
·
σ′′
∈
L
∧∀
e∈σπac (e)
∈Ai
∧
las (σ ′)/
∈
Ai
∧
i s (σ ′′)/
∈
Ai}
. Mo eo e , o all
σ∈L
, i
i s (σ ) /∈A1
, hen
L1=L1⊎ [��]
. Fu he mo e, o all
σ∈L
, i
las (σ) /∈A1
, hen
L1=L1⊎ [��]
. Also, i
|
L1|−|
j∈{2,...,n}
Lj| �=
1
,
L1=L1⊎ [��
|

j∈{2,...,n}Lj|−|L1|+1
]
.
Gi en a log
L= [�ab,b�]
and cu
(×,{a},{b})
, he esul a e sub-logs
L1= [�a�,��]
and
L2= [�b�]
.
As wi h he IM (Leemans e al. 2013b), he emp y aces all- h ough is applied i
enough emp y aces occu in an e en log. I he e a e no enough emp y aces o ig-
ge he all- h ough, hey a e emo ed om he e en log. Consequen ly, emp y aces
a e ne e p ojec ed on an ac i i y pa i ion.
E alua ion
In his sec ion, we e alua e ou p oposed me hod by conduc ing a ious expe imen s.
Fi s , we explain ou expe imen design, including he gene a ion p ocess o a anslu-
cen e en log. Second, we ocus on he esul s o he expe imen s ela ed o he sepsis
e en log (Mannha d 2016). Thi d, we ocus on he esul s o he expe imen s ela ed o
he oad a ic ine managemen e en log (de Leoni and Mannha d 2015). We imple-
men ed ou app oaches in Py hon using PM4Py (Be i e al. 2023)2.
i s (σ ) =

⊥, i σ= ��
πac (σ1), i σ�= ��
las
(σ ) =

⊥, i σ= ��
πac (σ|σ|), i σ�= ��
Fig. 10 O e iew o c ea ing anslucen e en logs using he IM (Leemans e al. 2013b)
2 Ou code is p o ided he e: h ps:// gi hub. com/ hhe b e b/ T ans lucen Ac i i yR ela i onshi ps
Page 19 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
Expe imen al se up
In his sec ion, we explain ou expe imen al se up. Fi s , we ocus on he gene a ion o
anslucen e en logs. Second, we explain how we measu e whe he ou disco e y ech-
niques imp o e exis ing echniques.
As discussed in Beyel and an de Aals (2022) and Beyel e al. (2024), enabled ac i i-
ies can be ex ac ed om sc eensho s using compu e - ision echniques. Howe e , we
gene a e a i icial anslucen e en logs o assess a ious scena ios and imp o e da a
a ailabili y and ep oducibili y. This me hod ollows a simila app oach o ha p esen ed
in Beyel and an de Aals (2022). An o e iew o he p ocess is shown in Fig.10. As
shown, a p ocess model is disco e ed om an exis ing e en log. The model and he
log a e aligned, and by conside ing only i ing aces, we inco po a e in o ma ion on
enabled ac i i ies using he eplay s a e in he model. Fo u he de ails, we e e o
Beyel and an de Aals (2022). Fo his, we use Pe i ne s disco e ed by he IM (Lee-
mans e al. 2013b) wi h di e en noise h esholds: 40 %, 60 %, and 80 %. We op ed no
o use he s anda d IM (Leemans e al. 2013a), as his algo i hm gene a es models ha
esul in anslucen e en logs wi h an excessi e numbe o enabled ac i i ies pe e en .
The chosen noise h esholds balance he p ep ocessing ime equi ed o c ea e syn-
he ic anslucen e en logs. A lowe h eshold would ha e made he p ocess oo ime-
consuming wi hou p o iding addi ional aluable insigh s. The ou pu o he IM is a
block-s uc u ed p ocess model. Consequen ly, he esul ing logs will cap u e his bias.
When disco e ing p ocess models om he c ea ed anslucen e en logs, he IM and
i s anslucen a ia ions (IM o, IM , and IM s) use a h eshold o 20 %, 40 %, 60 %,
and 80 %. We also in es iga ed whe he he e a e di e ences i a anslucen DFG o
classical DFG is used in all- h oughs, bu we did no obse e any di e ences wi hin ou
expe imen s.
To assess whe he anslucen ac i i y ela ionships add alue o p ocess disco e y,
we compa e he s anda d IM and IM , which do no accoun o enabled ac i i ies,
he anslucen IM (Beyel and an de Aals 2024a), i.e., IM o, IM , and IM s, and he
anslucen IM , i.e., IM o, IM , and IM s, p esen ed in his wo k. Addi ionally, we
e alua e whe he app oaches u ilizing in o ma ion on enabled ac i i ies disco e sim-
ila o imp o ed models wi h ewe a ian s compa ed o he o iginal algo i hms. To
explo e hese aspec s, we design e alua ion scena ios, as ou lined in Fig.11. We spli
Fig. 11 O e iew o ou e alua ion concep
Page 20 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
a anslucen e en log in o se e al sublogs, by ocusing on a ian s o he anslucen
e en log. The i s sublog includes all aces om he mos equen a ian , he second
sublog adds aces om he second mos equen a ian , and so on. Fo each sublog,
he IM (Leemans e al. 2013a), IM (Leemans e al. 2013b), he anslucen IM (Beyel
and an de Aals 2024a) (IM o, IM , and IM s), and ou new app oach (IM o, IM ,
and IM s) disco e p ocess models. To de e mine whe he a mo e ep esen a i e p o-
cess model is disco e ed wi h less in o ma ion, we compu e i ness, p ecision, and F1
sco es using he unspli anslucen e en log. Fi ness sco es a e calcula ed using align-
men s wi h a s anda d cos unc ion (Ad iansyah 2014). Fo p ecision sco es, we employ
a modi ied e sion o he escaping a cs app oach (Munoz-Gama and Ca mona 2010),
which conside s bo h he execu ed ac i i ies and he enabled ac i i ies in he log as
beha io . Only i ing aces a e analyzed. See Beyel and an de Aals (2024d) o mo e
in o ma ion. The F1 sco e is calcula ed as he ha monic mean o i ness and p ecision.
We also measu e simplici y by coun ing he numbe o places, ansi ions, and a cs in
each disco e ed Pe i ne .
Sepsis
Ou modi ied sepsis e en log (Mannha d 2016) can be ound in Beyel and an de Aals
(2024c). The h esholds 60 % and 80 % o he IM (Leemans e al. 2013b) o disco e he
model used as inpu o adding enabled ac i i ies esul ed in no di e ence in he pe o -
mance o he a ious algo i hms. Also, he h eshold o disco e ing p ocess models on he
gene a ed logs (see De ini ions14, 15, 16, 17 and 18) had a minimal impac . The e is only
a di e ence in pe o mance when compa ing he classical IM se ing o o he alues. The
esul s a e depic ed in Fig.12. The esul s indica e ha he anslucen a ian s achie e
pe ec i ness wi h ewe a ian s while s ill p o iding pe ec p ecision. This highligh s
ha ou me hods disco e p ocess models ha can be used o ackle he gene aliza ion
issue since ewe a ian s a e needed o disco e a well- ep esen a i e p ocess model.
When looking a he anslucen disco e y algo i hms, we can deno e ha all simul-
aneously achie e a pe ec i ness sco e. Howe e , he models disco e ed by IM o and
IM o i be e when ewe a ian s a e conside ed. Conside ing simplici y, he ans-
lucen a ia ions a e mo e s able and beha e simila ly o he di e en a ian coun s.
The inal models o he non- anslucen disco e y algo i hms a e sligh ly simple .
When using a h eshold o 40 % o he IM (Leemans e al. 2013b) o disco e ing
he model ha is used o anno a e he classic e en logs, we obse e simila esul s.
The esul s a e depic ed in Fig.13. As we deno e, he disco e y algo i hms using
in o ma ion on enabled ac i i ies pe o m be e . We can obse e ha i he h eshold
o disco e ing models (desc ibed in De ini ions14, 15, 16, 17 and 18) is 40 % o 20 %,
pe ec i ness is achie ed by including mo e a ian s han wi h a highe h eshold. As
be o e, he p ecision o he models disco e ed by anslucen disco e y algo i hms is
be e han hose c ea ed by disco e y algo i hms ha do no use he in o ma ion on
enabled ac i i ies. Simplici y beha es simila ly o he p e ious expe imen .

Page 21 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
Road a ic ine managemen
Ou modi ied oad a ic ine managemen e en log (de Leoni and Mannha d 2015)
can be ound in Beyel and an de Aals (2024b). When he h eshold o he IM (Lee-
mans e al. 2013b) o disco e ing he g ound u h model is se o 80 % o 60 %, only
wo a ian s a e e u ned. Since such a low numbe gi es limi ed insigh s, we do no
ocus on he esul s o hese expe imen s.
When conside ing he anslucen e en log c ea ed wi h a model disco e ed by he
IM (Leemans e al. 2013b) wi h a h eshold o 40 %, we can obse e mo e di e ences
be ween he di e en h esholds o he IM and i s anslucen a ian s, IM o, IM ,
and IM s (see Fig.14). By dec easing he h eshold o disco e ing models om 80
% o 60 % (desc ibed in De ini ions14, 15, 16, 17 and 18), we allow o disco e ing
models wi h pe ec i ness o a la ge numbe o a ian s. By dec easing his h esh-
old o 40 %, we obse e ha he pe o mance o IM o and IM imp o es. By u he
dec easing his h eshold o 20 %, we obse e ha he IM s is able o disco e pe -
ec ly i ing models wi h a la ge numbe o a ian s. We can also deno e ha he IM
(Leemans e al. 2013b) pe o mance does no change o he di e en h eshold al-
ues. The models disco e ed by u ilizing in o ma ion on enabled ac i i ies a e simple .
The esul s o he IM (Leemans e al. 2013a), IM o, IM , and IM s (Beyel and an
de Aals 2024a) a e showcased in Fig.15. The pe o mance o he anslucen a i-
an s is ge ing close o each o he . Conce ning simplici y, he anslucen a ian s
o he IM usually ind again simple models. O e all, i seems ha IM and IM a e
p o iding he bes esul s.
Fig. 12 Resul s o he algo i hms on he modi ied sepsis e en log (Beyel and an de Aals 2024c). The
models used o anno a ing he e en logs we e disco e ed by he IM (Leemans e al. 2013b) wi h a noise
h eshold o 60 % and 80 %
Page 22 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
An o e iew o di e en p ocess models disco e ed by he app oaches p esen ed in
his wo k is gi en in Fig.16. The model disco e ed by he IM (see Fig.16a) allows o
a lo o pa allel beha io , hus lowe ing p ecision. The model c ea ed by he IM o (see
Fig.16b) allows o less pa allel beha io bu s ill has a pa allel block in a b anch. The
model gene a ed by he IM (see Fig.16c) eplaces he pa allel block wi h a decision.
The model disco e ed by he IM s (see Fig.16d) p o ides no pa allelism o choice in
he b anch in ques ion. Hence, he model has he highes p ecision sco e. The models
suppo he sco es desc ibed in Fig.14d. Also, he models show he alue o enabled
ac i i ies in p ocess disco e y.
Fig. 13 Resul s o he algo i hms on he modi ied sepsis e en log (Beyel and an de Aals 2024c). The
model used o anno a ing he e en log was disco e ed by he IM (Leemans e al. 2013b) wi h a noise
h eshold o 40 %
Page 23 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
Fig. 14 Resul s o he IM (Leemans e al. 2013a) and IM o, IM , and IM s p esen ed in his wo k on he
modi ied oad a ic ine managemen e en log (Beyel and an de Aals 2024b). The model used o
anno a ing he e en log was disco e ed by he IM (Leemans e al. 2013b) wi h a noise h eshold o 40 %
Page 24 o 28
Beyeland an de Aals P ocess Science (2025) 2:15
Discussion
In gene al, wo king wi h anslucen e en da a ackles wo challenges desc ibed in
an de Aals e al. (2011): dealing wi h complex e en logs wi h di e se cha ac e is-
ics and balancing quali y c i e ia. Especially he gene aliza ion issue (Ca mona e al.
2018) can be ackled by u ilizing hese da a.
Fig. 15 IM (Leemans e al. 2013b) and i s anslucen a ian s (IM o, IM , and IM s (Beyel and an de Aals
2024a)) on he modi ied oad a ic ine managemen e en log (Beyel and an de Aals 2024b). The model
used o anno a ing he e en log was disco e ed by he IM (Leemans e al. 2013b) wi h a noise h eshold o
40 %
Fig. 16 Models disco e ed by IM and i s anslucen a ian s (IM o, IM , and IM s) on he comple e
modi ied oad a ic ine managemen e en log (Beyel and an de Aals 2024b). The model used o
anno a ing he e en log was disco e ed by he IM (Leemans e al. 2013b) wi h a noise h eshold o 40 %