Abstract
Project duration overruns remain a persistent challenge in the construction industry, leading to increased costs, compromised quality, environmental impacts, and reduced stakeholder confidence. In the UK alone, housing projects experienced average time overruns of 43% (2007–2019), and infrastructure projects saw a 65% increase in delivery timelines (2012–2021). Despite the evolution of scheduling tools such as CPM and PERT, embedded in platforms like Microsoft Project, Asta Powerproject and Primavera P6, their reliance on subjective input and limited data integration hinders their effectiveness, particularly at early project stages. Moreover, recent AI-enabled platforms like Procore, Autodesk Construction Cloud (ACC) remains focused on delivery stage planning and control of projects. This study, therefore aims to develop an automated time-estimating model that helps clients make informed decisions about project duration estimates especially at the early stages of the project life cycle.
This study critically reviews existing scheduling practices and proposes a data-driven alternative: the Intelligent Project Duration Estimating System (IPDES). Developed using a hybrid methodology combining Design Science Research, System Design, and traditional scientific paradigms, IPDES integrates 500 historical UK residential project datasets with categorised delay risks - client, contractor, consultant, and external related risks. The system uses Artificial Neural Networks (ANNs) in Neural designer to model non-linear relationships and prioritise stable input variables (e.g., internal floor area, number of units) over traditionally used but less stable predictors such as cost. The model underwent extensive training, input optimisation, and validation. The optimum ANN model achieved a normalised squared error (NSE) of 0.007, a coefficient of determination (R²) of 0.993, and an average prediction error of 0.51%, significantly outperforming baseline models. Error distribution analysis revealed that over 30% of predictions clustered around zero error, and predicted durations more closely matched actual completions than initial planned estimates, highlighting the model’s superior forecasting and generalisability reliability.
Beyond addressing structural inefficiencies and fragmented planning practices, the research presents IPDES as a scalable, intelligent solution for early-stage project duration forecasting. The study’s findings challenge cost-centric estimation methods and reinforce the value of AI in enhancing predictive precision and operational efficiency in construction project management. The research concludes with forwardlooking recommendations, including: establishing a centralised industry repository for historical and risk data; developing integrated, project-based risk management systems; and promoting user-friendly, cost-effective AI platforms accessible to nontechnical users. Future research should explore sub-duration predictions (e.g., piling, substructure) and plug-in integration with scheduling tools like Asta Powerproject and Oracle Primavera. These initiatives aim to foster widespread adoption of intelligent systems, driving more efficient, risk-aware, and data-centric construction project delivery.
This study critically reviews existing scheduling practices and proposes a data-driven alternative: the Intelligent Project Duration Estimating System (IPDES). Developed using a hybrid methodology combining Design Science Research, System Design, and traditional scientific paradigms, IPDES integrates 500 historical UK residential project datasets with categorised delay risks - client, contractor, consultant, and external related risks. The system uses Artificial Neural Networks (ANNs) in Neural designer to model non-linear relationships and prioritise stable input variables (e.g., internal floor area, number of units) over traditionally used but less stable predictors such as cost. The model underwent extensive training, input optimisation, and validation. The optimum ANN model achieved a normalised squared error (NSE) of 0.007, a coefficient of determination (R²) of 0.993, and an average prediction error of 0.51%, significantly outperforming baseline models. Error distribution analysis revealed that over 30% of predictions clustered around zero error, and predicted durations more closely matched actual completions than initial planned estimates, highlighting the model’s superior forecasting and generalisability reliability.
Beyond addressing structural inefficiencies and fragmented planning practices, the research presents IPDES as a scalable, intelligent solution for early-stage project duration forecasting. The study’s findings challenge cost-centric estimation methods and reinforce the value of AI in enhancing predictive precision and operational efficiency in construction project management. The research concludes with forwardlooking recommendations, including: establishing a centralised industry repository for historical and risk data; developing integrated, project-based risk management systems; and promoting user-friendly, cost-effective AI platforms accessible to nontechnical users. Future research should explore sub-duration predictions (e.g., piling, substructure) and plug-in integration with scheduling tools like Asta Powerproject and Oracle Primavera. These initiatives aim to foster widespread adoption of intelligent systems, driving more efficient, risk-aware, and data-centric construction project delivery.
| Original language | English |
|---|---|
| Qualification | Doctor of Philosophy |
| Awarding Institution |
|
| Supervisors/Advisors |
|
| Award date | 1 Apr 2025 |
| Publisher | |
| Publication status | Published - 1 Apr 2025 |