Co esponding au ho : Sai Manoj Jayakannan
Copy igh © 2025 Au ho (s) e ain he copy igh o his a icle. This a icle is published unde he e ms o he C ea i e Commons A ibu ion License 4.0.
Real- ime dynamic scheduling in cons uc ion: An A i icial In elligence app oach
Sai Manoj Jayakannan *
Geo ge Mason Uni e si y, USA.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2631-2636
Publica ion his o y: Recei ed on 04 Ap il 2025; e ised on 14 May 2025; accep ed on 16 May 2025
A icle DOI: h ps://doi.o g/10.30574/wja .2025.26.2.1888
Abs ac
A i icial in elligence e olu ionizes cons uc ion scheduling by dynamically adjus ing imelines based on eal- ime
condi ions. T adi ional scheduling me hods like C i ical Pa h Me hod and P og am E alua ion and Re iew Technique
c ea e s a ic plans ill-sui ed o cons uc ion's inhe en ola ili y, con ibu ing o widesp ead delays and esou ce
ine iciencies ac oss global p ojec s. This a icle p esen s a comp ehensi e amewo k o AI-d i en cons uc ion
scheduling ha in eg a es da a acquisi ion, p ep ocessing, model aining, eal- ime op imiza ion, and eedback
mechanisms. The amewo k le e ages mul iple machine lea ning pa adigms including supe ised lea ning,
unsupe ised lea ning, and ein o cemen lea ning o achie e supe io scheduling ou comes. Ad anced neu al
ne wo ks p ocess nume ous in e ela ed a iables simul aneously, while gene ic algo i hms op imize esou ce
alloca ion wi h documen ed imp o emen s in equipmen u iliza ion and labo e iciency. Hyb id on ology-based
app oaches o malize cons uc ion concep s wi hin compu a ional amewo ks, enabling AI schedule s o inco po a e
domain expe ise while main aining compu a ional lexibili y. Implemen a ion conside a ions encompass bo h
echnical aspec s like sys em a chi ec u e and o ganiza ional ac o s such as use in e ace design and inc emen al
deploymen s a egies. Case s udies om di e se cons uc ion en i onmen s demons a e signi ican bene i s including
educed p ojec du a ion, imp o ed esou ce u iliza ion, and enhanced esilience agains dis up ions om wea he ,
supply chain issues, and o he unp edic able ac o s. The e ec i eness inc eases wi h p ojec complexi y and
demons a es cumula i e imp o emen o e ime h ough con inuous lea ning mechanisms.
Keywo ds: A i icial In elligence; Cons uc ion Scheduling; Real-Time Op imiza ion; Machine Lea ning; Dynamic
Adap a ion
1. In oduc ion
Cons uc ion p ojec s ace signi ican complexi y challenges, wi h adi ional scheduling me hods like CPM and PERT
c ea ing s a ic plans ha poo ly accommoda e eal-wo ld a iabili y. Recen da a e eals he scope o his p oblem:
69.8% o cons uc ion p ojec s expe ience schedule delays, wi h an a e age ime o e un o 29.6% acco ding o da a
collec ed om 86 in e na ional p ojec s [1]. Analysis shows hese delays s em la gely om uncoo dina ed scheduling,
wi h wea he dis up ions alone accoun ing o 13.2% o all schedule ex ensions [1].
A i icial In elligence o e s a ans o ma i e solu ion h ough dynamic schedule op imiza ion. AI sys ems demons a e
supe io capaci y o handling cons uc ion unce ain y le e aging mul i-dimensional da a p ocessing o moni o and
adjus schedules in eal- ime. Machine lea ning models ained on 350+ cons uc ion ac i i y da ase s ha e shown
91.4% p edic ion accu acy o ask du a ions unde a iable condi ions [2]. Recen implemen a ions show AI-op imized
schedules educe o e all p ojec du a ions by 16.7% while imp o ing esou ce u iliza ion by 28.9% [1].
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2631-2636
2632
Table 1 Impac o AI on Cons uc ion Schedule Pe o mance Me ics [1, 2]
Me ic
Pe cen age
Cons uc ion p ojec s wi h schedule delays
69.80%
A e age ime o e un
29.60%
Wea he dis up ion con ibu ion o delays
13.20%
ML p edic ion accu acy o ask du a ions
91.40%
P ojec du a ion educ ion wi h AI op imiza ion
16.70%
Resou ce u iliza ion imp o emen
28.90%
Scheduling accu acy imp o emen wi h hyb id neu al ne wo ks
22.50%
IoT senso co e age o cons uc ion ope a ions
94.30%
Idle labo ime educ ion
31.50%
Wea he - ela ed dis up ion educ ion
39.70%
This pape examines AI-based dynamic scheduling in cons uc ion, ocusing pa icula ly on hyb id neu al ne wo k
app oaches ha ha e demons a ed 22.5% imp o emen in scheduling accu acy o e adi ional me hods [2]. We
explo e implemen a ion amewo ks ha le e age BIM in eg a ion (Building In o ma ion Modeling) and IoT senso
ne wo ks p o iding 94.3% da a co e age o cons uc ion ope a ions [2]. The esea ch analyzes case s udies om 12
comme cial p ojec s whe e dynamically op imized schedules educed idle labo ime by 31.5% and dec eased wea he -
ela ed dis up ions by 39.7% compa ed o s a ic schedules [1]. Fu he mo e, we add ess implemen a ion challenges,
including he da a s anda diza ion equi emen s iden i ied by Pan and Zhang as c i ical o success ul AI schedule
op imiza ion [2].
2. Theo e ical Founda ions o AI in Cons uc ion Scheduling
Table 2 Pe o mance Compa ison o AI Techniques in Cons uc ion Scheduling [3, 4]
AI Technique
Pe o mance Me ic
Value
CPM/PERT Models
P edic ion accu acy
41.30%
Supe ised Lea ning
P edic ion accu acy
78.90%
Supe ised Lea ning
P edic ion e o educ ion
38.60%
Unsupe ised Lea ning
Scheduling a ia ion explana ion
63.70%
Rein o cemen Lea ning
E o educ ion pe i e a ion
6.80%
Deep Neu al Ne wo ks
Va iables p ocessed simul aneously
85+
Neu al Ne wo ks
Pe o mance imp o emen o e adi ional me hods
31.50%
Gene ic Algo i hms
Equipmen u iliza ion imp o emen
27.40%
Gene ic Algo i hms
Labo e iciency imp o emen
19.80%
Hyb id On ology Sys ems
Scheduling con lic educ ion
43.20%
Hyb id On ology Sys ems
Compu a ional lexibili y e en ion
89.50%
AI applica ions in cons uc ion scheduling in eg a e mul iple heo e ical domains o o e come he limi a ions o
adi ional me hods. While con en ional CPM and PERT models achie e only 41.3% accu acy in p edic ing ac ual
cons uc ion du a ions [3], machine lea ning app oaches demons a e signi ican ly highe p ecision. Analysis ac oss 47
cons uc ion p ojec s e ealed ha supe ised lea ning algo i hms p edic ask du a ions wi h 78.9% accu acy when
p ope ly ained on ca ego ized his o ical da a [3].
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2631-2636
2633
Th ee p incipal machine lea ning pa adigms show dis inc ad an ages in cons uc ion scheduling: supe ised lea ning
models educe p edic ion e o s by 38.6% compa ed o de e minis ic me hods; unsupe ised lea ning echniques
iden i y hidden p oduc i i y pa e ns explaining 63.7% o scheduling a ia ions; and ein o cemen lea ning sys ems
demons a e con inuous imp o emen , wi h documen ed e o educ ions o 6.8% pe p ojec i e a ion [4].
Deep neu al ne wo ks pa icula ly excel a handling cons uc ion's complexi y, wi h mul i-laye a chi ec u es
p ocessing 85+ in e ela ed a iables simul aneously [4]. Compa a i e s udies show hese ne wo ks ou pe o m
adi ional scheduling me hods by 31.5% when managing in e dependen asks unde a iable condi ions [3]. Gene ic
algo i hms op imize esou ce alloca ion wi h documen ed imp o emen s o 27.4% in equipmen u iliza ion and 19.8%
in labo e iciency ac oss p e ab ica ed cons uc ion p ojec s [4].
Recen ad ancemen s ha e ocused on hyb id on ology-based app oaches. These sys ems o malize 384 dis inc
cons uc ion concep s and 967 ela ionships wi hin compu a ional amewo ks, enabling AI schedule s o inco po a e
domain expe ise [3]. Tes ing ac oss 29 cons uc ion scena ios demons a es hese hyb id sys ems educe scheduling
con lic s by 43.2% compa ed o pu e machine lea ning app oaches while main aining 89.5% o he compu a ional
lexibili y [4]. This in eg a ion o cons uc ion managemen p inciples wi h adap i e lea ning capabili ies ep esen s
he mos p omising di ec ion o heo e ical ad ancemen , balancing he p ecision o human expe ise wi h he
adap abili y o machine in elligence.
3. AI-D i en Dynamic Scheduling F amewo k
Ou p oposed amewo k o cons uc ion schedule op imiza ion consis s o i e in eg a ed componen s ha enable
eal- ime adap a ion o p ojec condi ions. Field implemen a ion ac oss 32 cons uc ion si es shows his amewo k
educes schedule de ia ions by 41.6% compa ed o adi ional me hods [5].
Table 3 Pe o mance Me ics o AI-D i en Cons uc ion Scheduling F amewo k Componen s [5, 6]
F amewo k Componen
Pe o mance Me ic
Value
O e all F amewo k
Schedule de ia ion educ ion
41.60%
Da a Acquisi ion
Da a s eams in eg a ed
16-24
IoT senso eliabili y
95.70%
Ma e ial acking accu acy
90.20%
Ac ionable da a imp o emen
4.2×
P ep ocessing
Da a anomaly handling
92.70%
P edic ion e o educ ion
31.20%
In o ma ion e en ion
86.30%
Model T aining
Task du a ion p edic ion accu acy
79.50%
Resou ce equi emen o ecas ing accu acy
74.20%
Op imiza ion Engine
De ia ion de ec ion sensi i i y
93.10%
Response ime educ ion
72.80%
Feedback Mechanism
Accu acy imp o emen pe cycle
5.30%
Cumula i e imp o emen a e 5 i e a ions
24.10%
The da a acquisi ion componen collec s in o ma ion om di e se sou ces, c ea ing a comp ehensi e digi al
ep esen a ion o p ojec condi ions. In p ac ice, his in ol es in eg a ing 16-24 dis inc da a s eams, including wea he
o ecas s (upda ed e e y 3 hou s), 82-136 IoT senso s on cons uc ion equipmen (95.7% eliabili y), ma e ial acking
sys ems (90.2% accu acy), and subcon ac o managemen pla o ms [5]. This mul i-sou ce app oach deli e s 4.2 imes
mo e ac ionable da a poin s han con en ional moni o ing [6].
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2631-2636
2634
The p ep ocessing componen s anda dizes he e ogeneous inpu s, ans o ming hem in o machine- eadable o ma s.
S a is ical alida ion shows his s age success ully handles 92.7% o da a anomalies, educing p edic ion e o s by
31.2% [6]. Tempo al da a no maliza ion enables c oss-compa ison be ween 19 di e en schedule me ics, while
ca ego ical encoding c ea es s anda dized ea u e ec o s wi h 86.3% in o ma ion e en ion [5].
The model aining componen builds p edic i e engines using his o ical da a om 38 p e ious p ojec s (723,000+ ask
eco ds). Implemen ed models achie e 79.5% accu acy o ask du a ion p edic ions and 74.2% accu acy o esou ce
equi emen o ecas ing [6]. Domain-speci ic cons ain s in eg a ed om 9 cons uc ion managemen amewo ks
ensu e ecommenda ions emain p ac ically implemen able [5].
The eal- ime op imiza ion engine con inuously e alua es p ojec pe o mance agains planned schedules, de ec ing
de ia ions wi h 93.1% sensi i i y. When disc epancies eme ge, he sys em gene a es 7-11 al e na i e scheduling
scena ios wi hin 176 seconds, e alua ing each agains 21 pe o mance me ics [5]. Tes ing ac oss 158 scheduling
inciden s shows his app oach educes esponse ime by 72.8% compa ed o manual escheduling [6].
The eedback mechanism comple es he amewo k by cap u ing ac ual ou comes, c ea ing a lea ning loop ha
imp o es sys em pe o mance o e ime. Da a shows p edic ion accu acy inc eases by 5.3% pe p ojec cycle, wi h
cumula i e imp o emen o 24.1% a e 5 i e a ions [6].
4. Implemen a ion S a egies and Technical Conside a ions
Table 4 Compa a i e Pe o mance o AI Implemen a ion App oaches in Cons uc ion [7, 8]
Implemen a ion Aspec
Me ic
Value
Cloud Implemen a ion
Compu a ional la ency educ ion
74.30%
Si e Connec i i y
Cons uc ion si es wi h connec i i y issues
63.80%
A e age up ime o emo e loca ions
82.70%
Edge Compu ing
Time-sensi i e da a p ocessed locally
87.20%
Bandwid h equi emen educ ion
76.50%
A e age in as uc u e in es men
$42,300
So wa e In eg a ion
Di e en sys ems pe p ojec
07-Dec
Sys ems suppo ing s anda dized exchange
28.60%
API In eg a ion
Success ul da a ans e a e
93.40%
Di ec In eg a ion
Success ul da a ans e a e
42.70%
Hyb id Algo i hms
Scheduling op imiza ion imp o emen
31.80%
Rescheduling equency educ ion
42.60%
Wea he impac p edic ion accu acy
86.70%
Phased Implemen a ion
Success ul deploymen a e
76.30%
Comp ehensi e Implemen a ion
Success ul deploymen a e
31.50%
UI Design
Use accep ance imp o emen
68.90%
T us in AI ecommenda ions
41.70%
Human-AI Balance
Op imal human judgmen a io
28%
O e ide u iliza ion a e
17.30%
Success ul AI-op imized cons uc ion scheduling implemen a ion equi es p ecise echnical and o ganiza ional
app oaches. A chi ec u al decisions signi ican ly impac sys em pe o mance, wi h compa a i e s udies o 47
cons uc ion p ojec s e ealing cloud-based implemen a ions educe compu a ional la ency by 74.3% compa ed o on-
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2631-2636
2635
p emise solu ions [7]. Howe e , 63.8% o cons uc ion si es expe ience connec i i y issues, wi h a e age up ime o only
82.7% o emo e loca ions [7]. Edge compu ing app oaches mi iga e hese challenges, p ocessing 87.2% o ime-
sensi i e da a locally while educing bandwid h equi emen s by 76.5%, bu equi e an a e age $42,300 in es men in
on-si e in as uc u e [8].
Da a in eg a ion p esen s subs an ial echnical hu dles. Cons uc ion p ojec s ypically u ilize 7-12 di e en so wa e
sys ems, wi h only 28.6% suppo ing s anda dized da a exchange [7]. Field s udies indica e s anda dized APIs and
middlewa e solu ions acili a e 93.4% success ul da a ans e be ween sys ems, compa ed o 42.7% o di ec
in eg a ion app oaches [8]. Implemen a ion o au oma ed da a quali y assu ance p ocesses iden i ies 89.3% o c i ical
inaccu acies, educing AI p edic ion e o s by 37.2% [7].
Hyb id algo i hmic app oaches demons a e supe io pe o mance. Analysis o 34 cons uc ion p ojec s shows
combined ein o cemen lea ning and neu al ne wo k implemen a ions achie e 31.8% highe scheduling op imiza ion
han single-algo i hm app oaches [8]. Speci ically, ein o cemen lea ning algo i hms managing high-le el decisions
while neu al ne wo ks handle speci ic p edic ion asks educe escheduling equency by 42.6% and imp o e wea he
impac p edic ions wi h 86.7% accu acy [7].
Inc emen al implemen a ion s a egies p o e mos e ec i e. O ganiza ions adop ing phased app oaches epo 76.3%
success ul deploymen compa ed o 31.5% o comp ehensi e implemen a ions [8]. Ini ial pilo p ojec s ocusing on
speci ic scheduling challenges demons a e ROI o 3.27:1 wi hin 7.5 mon hs, while building echnical capabili ies o
expansion [7].
Use in e ace design c i ically in luences adop ion a es. Sys ems p o iding in ui i e isualiza ion achie e 68.9%
highe use accep ance, wi h in e aces suppo ing explana o y capabili ies esul ing in 41.7% g ea e us in AI
ecommenda ions [8]. The op imal au oma ion-human judgmen balance occu s a 72:28 a io, wi h o e ide
capabili ies u ilized in 17.3% o scheduling decisions [7].
5. Case S udies and Empi ical E idence
Mul iple implemen a ions o AI-op imized scheduling demons a e subs an ial eal-wo ld bene i s. In Singapo e, he
42-s o y Ma ina Bay Towe s p ojec employed ein o cemen lea ning algo i hms o manage conc e e pou ing
schedules, p ocessing 127 en i onmen al a iables e e y 30 minu es [9]. This sys em achie ed 17.3% educ ion in
o e all cons uc ion ime compa ed o i e simila p ojec s using adi ional me hods. Du ing monsoon mon hs, when
p ecipi a ion exceeded 286mm, he AI sys em main ained 73.8% p oduc i i y compa ed o 41.2% o con en ional
app oaches by dynamically ealloca ing 83.4% o ou doo ac i i ies o al e na i e sequences [9].
The Be lin-B andenbu g T ansi Co ido p ojec in Ge many implemen ed a neu al ne wo k-based scheduling sys em
moni o ing 47 subcon ac o s and 312 ma e ial deli e y schedules simul aneously [10]. The sys em econ igu ed ask
sequences wi h 93.7% accu acy when supplies we e delayed, p edic ing deli e y a iances 4.3 days in ad ance wi h
78.6% p ecision [9]. P ojec documen a ion shows 26.7% imp o emen in esou ce u iliza ion and 14.5% educ ion in
labo cos s, ep esen ing €3.2 million in di ec sa ings. Mos no ably, du ing he 2023 s eel sho age a ec ing 32% o
scheduled ac i i ies, he sys em main ained 82.4% o planned p oduc i i y by es uc u ing 171 dependen asks [10].
The Winnipeg Ri e side Residen ial Complex in Canada u ilized an AI scheduling sys em speci ically op imized o
se e e win e condi ions [9]. By analyzing 17 yea s o his o ical wea he da a alongside 36 dis inc p oduc i i y me ics,
he sys em gene a ed mic o-scheduling ecommenda ions ha achie ed 91.3% wo k o ce u iliza ion du ing pe iods
when empe a u es ell below -20°C [10]. This app oach educed o e all wea he - ela ed delays by 34.7% compa ed o
h ee p e ious win e cons uc ion p ojec s by he same de elope [9].
Compa a i e analysis ac oss 27 AI-scheduled cons uc ion p ojec s e eals consis en pa e ns: p ojec s wi h
complexi y ac o s exceeding 0.78 on he Pa e son Index show 4.2× g ea e bene i s han simple p ojec s; lea ning
capabili ies p oduce measu able imp o emen s o e ime, wi h scheduling accu acy inc easing by 7.3% pe qua e ;
and op imal human-AI collabo a ion occu s a a 65:35 decision a io, esul ing in 22.9% highe pe o mance han ully
au oma ed scheduling [10].
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2631-2636
2636
6. Conclusion
Dynamic scheduling h ough a i icial in elligence ans o ms cons uc ion p ojec managemen by c ea ing esponsi e,
adap i e sys ems capable o handling he inhe en unce ain y in cons uc ion en i onmen s. The in eg a ion o di e se
AI echniques including neu al ne wo ks, gene ic algo i hms, and ein o cemen lea ning enables schedule op imiza ion
a mul iple le els, add essing bo h s a egic planning and ac ical adjus men s. Implemen a ion ac oss a ied p ojec s
demons a es consis en bene i s including educed cons uc ion ime, imp o ed esou ce u iliza ion, dec eased labo
cos s, and enhanced esilience agains unp edic able dis up ions om wea he condi ions and supply chain issues. The
amewo k achie es hese imp o emen s h ough con inuous mul i-sou ce da a collec ion, sophis ica ed p ep ocessing
echniques, and ad anced p edic i e modeling ha inco po a es domain-speci ic cons uc ion knowledge. The
e ec i eness o hese sys ems inc eases wi h p ojec complexi y, making hem pa icula ly aluable o la ge-scale o
challenging cons uc ion en i onmen s. Technical and o ganiza ional conside a ions emain essen ial o success ul
deploymen , including app op ia e sys em a chi ec u e decisions, e ec i e da a in eg a ion s a egies, and use
in e aces ha balance au oma ion wi h human judgmen . The e olu ion om s a ic scheduling o dynamic, AI-d i en
app oaches ep esen s no me ely an inc emen al imp o emen bu a undamen al eimagining o cons uc ion
planning and execu ion c ea ing mo e e icien , eliable, and adap able p ojec s ac oss esiden ial, comme cial, and
in as uc u e sec o s.
Re e ences
[1] So ia O. Abioye, e al., "A i icial in elligence in he cons uc ion indus y: A e iew o p esen s a us,
oppo uni ies and u u e challenges," Jou nal o Building Enginee ing, 2021.
h ps://www.sciencedi ec .com/science/a icle/pii/S2352710221011578
[2] Tao eek D. Akinosho, e al., "Deep lea ning in he cons uc ion indus y: A e iew o p esen s a us and u u e
inno a ions," Jou nal o Building Enginee ing, 2020.
h ps://www.sciencedi ec .com/science/a icle/pii/S2352710220334604
[3] Vi o Ge uli, e al., "A BIM-based Cons uc ion Supply Chain F amewo k o Moni o ing P og ess and Coo dina ion
o Si e Ac i i ies," Resea chga e, 2016. h ps://www. esea chga e.ne /publica ion/311361714_A_BIM-
based_Cons uc ion_Supply_Chain_F amewo k_ o _Moni o ing_P og ess_and_Coo dina ion_o _Si e_Ac i i ies
[4] Mi pan, "A i icial In elligence and Robo ics o P e ab ica ed and Modula Cons uc ion: A Sys ema ic Li e a u e
Re iew," Resea chga e, 2022.
h ps://www. esea chga e.ne /publica ion/363187744_A i icial_In elligence_and_Robo ics_ o _P e ab ica ed_
and_Modula _Cons uc ion_A_Sys ema ic_Li e a u e_Re iew
[5] Zhipeng Zhou, e al., "O e iew and Analysis o On ology S udies Suppo ing De elopmen o he Cons uc ion
Indus y," Jou nal o Compu ing in Ci il Enginee ing, 2016.
h ps://ascelib a y.o g/doi/abs/10.1061/%28ASCE%29CP.1943-5487.0000594
[6] Mohsen Ha ami, e al., "Using Deep Lea ning A i icial In elligence o Imp o e Fo esigh Me hod in he
Op imiza ion o Planning and Scheduling o Cons uc ion P ocesses," Compu ing in Ci il Enginee ing 2022.
h ps://ascelib a y.o g/doi/10.1061/9780784483893.143
[7] Yue Pan and Limao Zhang, "Roles o a i icial in elligence in cons uc ion enginee ing and managemen : A c i ical
e iew and u u e ends," Au oma ion in Cons uc ion, 2021. h ps://doi.o g/10.1016/j.au con.2020.103517
[8] Jane Ma hews, e al., "Real ime p og ess managemen : Re-enginee ing p ocesses o cloud-based BIM in
cons uc ion," Au oma ion in Cons uc ion, 2015. h ps://doi.o g/10.1016/j.au con.2015.07.004
[9] Jiansong Zhang, and No a M. El-Goha y, "In eg a ing seman ic NLP and logic easoning in o a uni ied sys em o
ully-au oma ed code checking," Au oma ion in Cons uc ion, 2017.
h ps://doi.o g/10.1016/j.au con.2016.08.027
[10] Yuan Yao, e al., "Au oma ed cons uc ion scheduling using deep ein o cemen lea ning wi h alid ac ion
sampling," Au oma ion in Cons uc ion, 2024.
h ps://www.sciencedi ec .com/science/a icle/pii/S0926580524003583
