1. Home
  2. TRID
  3. View Record
https://www.ntro.org.au/

Exploring the use of artificial intelligence (AI) solutions to improve the accuracy of project delivery forecasts

The Queensland Department of Transport and Main Roads (TMR), in collaboration with the Australian Road Research Board (ARRB), has investigated the suitability and performance of artificial intelligence (AI) solutions to enhance decision making through improved accuracy and precision of project cost and duration forecasting. Machine learning (ML) models use reference class forecasting to identify predictive patterns in large datasets. With appropriate instructions, suitable data and calibration, ML models provide improved final cost forecasting accuracy that may assist TMR's budgeting capabilities for individual projects and their work program. The project scope entailed an evaluation of the suitability of TMR project data for use in ML models as well as the effectiveness and accuracy of ML models in predicting the forecast final costs and forecast completion dates of projects using provided datasets. Machine learning models use ample amounts of high-quality project data that have minimal missing values. A collection of 116 TMR projects that demonstrated high levels of adherence to machine learning suitability factors formed the basis of the model’s dataset. This data comprised 89 small (< $10m) and 26 large ($10m – < $100m) budget projects that encompassed 16 different work types and 15 delivery programs. Predictive analytics were carried out on TMR’s dataset, which presented evidence-based predictive and descriptive observations. Predictive observations were thereafter identified in the parameters of cost/duration forecasting accuracy and budget adjustment. The parameters of project capital expenditure, construction year, wet season start month, work type and delivery program were identified as descriptive observations and therefore cannot be applied to the overall TMR portfolio. Additionally, these analytics found that projects typically return capital budget and/or contingency late in the project life cycle. Machine learning modelling for project cost and duration forecasting was developed and tested. Improving the predictability of project cost and duration under/overruns is a challenging problem affecting major projects. Artificial intelligence technologies such as machine learning models offer a faster, cheaper, and more powerful way to conduct many thousands of experiments to build evidence-based predictive models for tackling these challenges. TMR’s substantial project database has demonstrated applicability for ML modelling to improve the accuracy and precision of project cost and duration forecasting. The ML model identified significant financial and delivery opportunities for the re-distribution of capital budget and/or contingency back into their portfolio, given that forecast project costs and contingency may be predicted earlier and more accurately. Overall, AI technologies have demonstrated capabilities that enhance capital productivity, portfolio performance, early warning capability and a decrease in monitoring costs. ML predictions may be a valuable and effective tool for Project and Program Managers to validate traditional cost and duration forecasts.

Language

  • English

Media Info

  • Pagination: 37p

Subject/Index Terms

Filing Info

  • Accession Number: 01773910
  • Record Type: Publication
  • Source Agency: ARRB Group Limited
  • Report/Paper Numbers: O18
  • Files: ATRI
  • Created Date: Jun 8 2021 12:24PM