Enhancing Construction Project Performance Using Artificial Intelligence Techniques
A validated, live-deployed framework integrating NLP extraction, entropy-weighted dynamic KPI prioritisation, and reinforcement learning — validated across five international case studies in four countries.
01 / Research Motivation
Construction Projects Fail Silently.By the time the report arrives, the damage is done.
Iraq's construction sector loses an estimated 30–40% of project budgets to delays, cost overruns, and reporting failures. The root cause is not resource scarcity — it is a fundamental gap in real-time performance intelligence.
Conventional monitoring relies on weekly manual reports. Critical signals — a schedule slip, a cost variance, a safety near-miss — go undetected for days. When they surface, corrective action is reactive. This research eliminates that gap.
Reporting Latency
Weekly manual reports delay issue detection by 5–7 days. By then, recovery costs multiples of prevention.
Unstructured Communication
Site teams communicate in mixed Arabic-English. No prior construction AI system addressed this code-switching at field level.
Static KPI Weighting
Traditional frameworks apply fixed weights regardless of project context, archetype, or evolving site conditions.
No Early Warning
Existing systems detect problems after thresholds are breached. IPAF detects the signal 30 days before the breach.
02 / Framework Architecture
Three Integrated AI Layers.One Continuous Performance Signal.
IPAF is the first construction performance framework to integrate NLP extraction, entropy-weighted dynamic KPI prioritisation, and reinforcement learning in a single operational system — validated on live international projects.
Natural Language Processing
Raw project communications in Arabic, English, or mixed language are processed by a self-hosted LLM (Qwen2.5-7B). Four extraction agents operate concurrently: KPI Extractor, Risk Detector, Message Summariser, and Sentiment Analyser. Confidence-scored outputs feed the weighting engine weekly.
Entropy-Weighted Dynamic KPI Prioritisation
α + β + γ = 1.0 · each ≥ 0.10 · adjusted weekly by RL agent
RL State Classification + Early Warning
A reinforcement learning agent classifies project health into one of six states weekly. Three early warning pathways fire 30 days before conventional thresholds are reached. Human oversight is mandatory at four defined trigger points.
03 / Empirical Validation
Five Projects. Four Countries.One Framework. Consistent Results.
Validated across five international construction projects using prospective, retrospective, and mixed methodologies. Across all five cases, the data-driven EWM coefficient (γ) increased — the primary cross-case finding.
BC Highway Infrastructure
46 Active KPIs
✓ 10/10 KPIs improved
Erbil Residential Complex
47 Active KPIs
✓ 10/10 KPIs improved · 30-day schedule warning
Baghdad Luxury High-Rise
41 Active KPIs
✓ EWM detected gradual decay vs binary suspension
Hamad General Hospital
37 Active KPIs
✓ 30-day strategic alignment warning (M5→M6)
Westgate Tunnel Melbourne
28 Active KPIs
✓ M3 safety entropy precursor before PFAS crisis
γ increased in all five cases regardless of starting archetype, country, or project type — confirming that accumulated project data consistently becomes the dominant weighting signal as projects mature.
04 / Live Deployment
Not a Prototype.A Deployed System.
IPAF is deployed on live infrastructure. The web application processes real project communications, extracts KPIs through a self-hosted LLM, runs the weighting and classification engine weekly, and serves role-differentiated dashboards to Project Managers, Site Engineers, and Clients.
05 / Research Contributions
Four Verified Contributionsto Construction AI Research
Dynamic KPI Taxonomy
121 KPIs from 218 studies, reduced to 57 through two-round Delphi (CV ≤ 0.30) across 10 unified domains. First construction KPI taxonomy combining expert consensus, strategic context, and data-driven entropy weighting in a single validated set.
Integrated AI Framework
First construction performance framework to integrate NLP extraction, entropy-weighted dynamic prioritisation, and reinforcement learning in a single operational system — validated on live international projects.
30-Day Early Warning
Three domain-specific pathways (schedule entropy, strategic alignment, safety entropy) demonstrated consistent 30-day lead time across CS2, CS4, and CS5 — before conventional reporting detected the same signals.
Cross-Case RL Evidence
γ coefficient increased across all five case studies regardless of archetype, country, or starting value — providing the first empirical evidence that EWM weighting consistently gains dominance as project data accumulates.
06 / دليل الاستخدام · User Guide
Getting Startedكيفية استخدام النظام
IPAF is designed for construction professionals at all levels of technical experience. Follow these steps to begin monitoring your project.
تم تصميم نظام IPAF للمهنيين في مجال البناء بجميع مستويات الخبرة التقنية. اتبع هذه الخطوات لبدء مراقبة مشروعك.
Open the system and enter your email and password. Your project manager will provide your credentials.
افتح النظام وأدخل بريدك الإلكتروني وكلمة المرور. سيزودك مدير المشروع ببيانات تسجيل الدخول.
Your dashboard shows the project state, KPI values, and any active alerts. The coloured badge at the top tells you the current project health.
تعرض لوحة التحكم حالة المشروع وقيم مؤشرات الأداء وأي تنبيهات نشطة. تشير الشارة الملونة في الأعلى إلى صحة المشروع الحالية.
Tap the chat button (bottom-right corner) and type your site update in Arabic or English. The system extracts KPI values automatically.
انقر على زر المحادثة (الزاوية اليمنى السفلية) واكتب تحديث الموقع بالعربية أو الإنجليزية. يستخرج النظام قيم مؤشرات الأداء تلقائياً.
Review the extracted values shown in the chat, then tap 'Save All' to update your project dashboard.
راجع القيم المستخرجة المعروضة في المحادثة، ثم انقر على 'حفظ الكل' لتحديث لوحة تحكم مشروعك.
07 / About
The Research & Researcher
This framework is the product of doctoral research in Construction Project Management at the University of Baghdad. It addresses the acute gap between modern AI capabilities and their adoption in Iraqi and MENA-region construction practice.
The live system — deployed on real infrastructure and validated across five international case studies — serves simultaneously as the dissertation's empirical contribution and a replicable methodology for integrated AI performance management in construction.
| Researcher | Zaid Khalid |
| Degree | PhD · Construction Management |
| Institution | University of Baghdad |
| Department | College of Engineering · Civil Engineering |
| Supervisor | Prof. Dr. Meervat R. Altaie |
| Year | 2026 |