Search
Article

Future of forecast: machine learning

Oğuzhan Çınar

Managing Consultant

oguzhancinar@hka.com

The impact of construction

The construction industry, which impacts and interacts with more than 200 sectors, is a significant contributor to the development, prosperity and standing of countries. The scale of construction – which now accounts for 13%[1]Ribeirinho, M. J. et al. (2020) ‘The next normal in construction’, Mckinsey & Company of global GDP and 8.6%[2]Mella, A. and Savage, M. (2018) ‘Construction sector employment in low income countries’ of global employment – together with the interdependency that the sector has on so many other parts of the economy, means that there is a direct correlation between the performance of construction and other market sectors.

Such interrelationships between sectors extend not only to performance, but of course also to operational working practices that are increasingly reliant on the use of improved technology. In spite of developments over the past couple of decades, the construction industry still lags behind other sectors in terms of its adoption of technology, and continues to rely on traditional work methods. By not readily adopting new innovation, the construction industry tends to hamper its own ability to operate as efficiently as it might with other sectors.


The importance of reliable information

It is essential for construction companies – often engaged to deliver complex projects, which are inherently high risk, yet which often only yield low net profit margins – to consistently manage and monitor completion dates, anticipated final cost and risks that will impinge on delivery. To keep these elements on track relies, inevitably, on receiving and using accurate information from internal and external stakeholders. Even when working with stakeholders across the supply chain to track delivery performance through the use of assumptions, forecasts, and studies, such information inevitably remains susceptible to human error. Noting that 80%[3]MACE (2019), ‘A blueprint for modern infrastructure delivery’ of all large construction projects experience some form of delay and cost-overruns, this must at least in part, bring into question the accuracy of such monitoring and forecasting, when it is based on human input alone.

The potential of machine learning technology

So is there an alternative to the reliance on human input alone? One solution may be the use of Machine Learning Technology (“MLT”), which is one of the branches of Artificial Intelligence. In simple terms, MLT analyses historic data, learns from it and uses the trends identified to predict future events. To do so in an accurate and successful manner, MLT relies on being able to process a large volume of high quality datasets thereby enhancing the learning curve of the model. To provide some context, while the accuracy rate of an algorithm informed by 100 data sets is 50%, the accuracy of the same algorithm can reach up to 90% if 10,000 data sets are relied upon[4]Sessions, Valerie & Valtorta, Marco. (2006). ‘The Effects of Data Quality on Machine Learning Algorithms’. Where high quality large datasets could be deployed, recent studies and research[5]Pan, Y. and Zhang, L. (2021) ‘Roles of artificial intelligence in construction engineering and management: A critical review and future trends’, Automation in Construction, 122(October 2020), p. … Continue reading demonstrate that MLT could be used frequently by contractors, employers, and members of the construction supply chain to bring greater accuracy to project reporting. By extension, MLT could contribute to achieving greater profitability, productivity and efficiency of projects where used in concert with disciplines such as high calibre project management disciplines, planning, and health and safety management. Increasingly, this emerging technology is being seen as a tool that can be used as a support mechanism by companies to assist with decision making, forecasting and controlling their studies[6]Fitzsimmons, J., Hong, Y., & Brilakis, I. (2020) ‘Improving Construction Project Schedules before Execution’ Proceedings of the 37th International Symposium on … Continue reading.

We have witnessed that 80% of projects experiencing cost overruns[7]Flyvbjerg, B. (2014), ‘What You Should Now About Megaprojects, and Why’, Project Management Journal, pp. 6-19.doi:10.1002/pmj. Equally, recent studies demonstrate that MLT algorithms can, from an early stage, predict the total cost of the projects with over 90%[8]Ahiaga-Dagbui, D. D. and Smith, S. D. (2014) ‘Dealing with construction cost overruns using data mining’, Construction Management and Economics, 32(7–8), pp. 682–694. doi: … Continue reading accuracy. This, therefore, suggests a strong case for using such technology to assess project costs to a greater degree. In the instance of earthworks activity, for example, variables such as soil characteristics, excavated area, depth of overburden and dumping area, are calculated and the model generates possible total cost scenarios[9]Petroutsatou, K. et al. (2012) ‘Early Cost Estimating of Road Tunnel Construction Using Neural Networks’, Journal of Construction Engineering and Management, 138(6), pp. 679–687. doi: … Continue reading. This provides insight as to the most cost effective excavation method. Multiple examples of such models could be readily drawn together and processed by an MLT algorithm to verify them, and / or provide alternative costings calculations thereby generating more reliable forecasts underpinning the chosen excavation method.

Furthermore, as is widely accepted, there is a direct correlation between cost-overruns and delays in project completion. Project delay is one of the major challenges for companies and generally leads to claims, disputes, contract terminations, and puts reputations at stake. Therefore, forecasting a likely completion date accurately, predicated on likely delay events, is crucial for all parties[10]Gondia, A. et al. (2020) ‘Machine Learning Algorithms for Construction Projects Delay Risk Prediction’, Journal of Construction Engineering and Management, 146(1), p. 04019085. doi: … Continue reading. Whilst this is a good aim, the uniqueness of each project makes it difficult to always foresee future delay events. To address this, MLT algorithms seek to assess the reliability of baseline programmes by forecasting the likely finish dates, and possible delay factors that should have been considered before the commencement date. The model analyses data from previous projects and considers numerous delay factors related to an employer, contractor, other stakeholders, resources, and external factors. It then presents the likely factors that might go on to cause delay to the project. The current literature illustrates that the accuracy of MLT models on planning and delay management can be up to 90% accurate[11]Yaseen, Z. M. et al. (2020) ‘Prediction of risk delay in construction projects using a hybrid artificial intelligence model’, Sustainability (Switzerland), 12(4), pp. 1–14. doi: … Continue reading.

Another application to which MLT could potentially be applied in the construction sector is that of pre-emptively identifying health and safety risks. The construction industry is recognised as one of the most dangerous sectors, with the UK Health and Safety Executive estimating the cost of work-related injuries and illness in construction industry at £1.2 billion in 2019[12]Health and Safety Executive (2020), ‘Construction statistics in Great Britain’. Studies demonstrate that artificial neural network based models informed by over 90,000 health and safety reports provide detailed predictions of incident and injury types with more than 90%[13]Baker, H., Hallowell, M. R. and Tixier, A. J. P. (2020) ‘AI-based prediction of independent construction safety outcomes from universal attributes’, Automation in Construction, 118(June), p. … Continue reading accuracy. Companies could therefore mitigate possible health and safety related events and take precautions to prevent possible loss by deploying MLT algorithms on a more regular basis.

In summary

During a construction project, numerous assumptions and forecasts are generated and based on these studies stakeholders determine their short- and long-term strategies. However, such studies are open to human error. Therefore, an accurate forecast is invaluable. In the light of recent studies and developments, it has been seen that the use of MLT has already had a positive several sectors and could be used in a similar manner in the construction industry. Its success, however, relies on the use of a large volume of high quality datasets. For it to take root in construction will therefore require not only significant investment, but also a determination by construction practitioners to record and standardise project data diligently.


About the author

Oğuzhan Çınar has 5 years of construction industry experience as a project control and cost engineer working on billion-dollar transportation, infrastructure and building projects under FIDIC and bespoke forms of contract in Qatar, Tanzania and Turkey. He is skilled in project control and commercial management disciplines, including earned value analysis, activity and source-based budgeting, planning, claim and global settlement issues, as well as stakeholder and subcontractor management areas. Combining his knowledge in civil engineering with his experience in different projects and countries enables him to analyse events accurately and to understand their root causes.

References

References
1 Ribeirinho, M. J. et al. (2020) ‘The next normal in construction’, Mckinsey & Company
2 Mella, A. and Savage, M. (2018) ‘Construction sector employment in low income countries’
3 MACE (2019), ‘A blueprint for modern infrastructure delivery’
4 Sessions, Valerie & Valtorta, Marco. (2006). ‘The Effects of Data Quality on Machine Learning Algorithms’
5 Pan, Y. and Zhang, L. (2021) ‘Roles of artificial intelligence in construction engineering and management: A critical review and future trends’, Automation in Construction, 122(October 2020), p. 103517. doi: 10.1016/j.autcon.2020.103517.
6 Fitzsimmons, J., Hong, Y., & Brilakis, I. (2020) ‘Improving Construction Project Schedules before Execution’ Proceedings of the 37th International Symposium on Automation and Robotics in Construction (ISARC) https://doi.org/10.22260/isarc2020/0157
7 Flyvbjerg, B. (2014), ‘What You Should Now About Megaprojects, and Why’, Project Management Journal, pp. 6-19.doi:10.1002/pmj
8 Ahiaga-Dagbui, D. D. and Smith, S. D. (2014) ‘Dealing with construction cost overruns using data mining’, Construction Management and Economics, 32(7–8), pp. 682–694. doi: 10.1080/01446193.2014.933854.
9 Petroutsatou, K. et al. (2012) ‘Early Cost Estimating of Road Tunnel Construction Using Neural Networks’, Journal of Construction Engineering and Management, 138(6), pp. 679–687. doi: 10.1061/(asce)co.1943-7862.0000479
10 Gondia, A. et al. (2020) ‘Machine Learning Algorithms for Construction Projects Delay Risk Prediction’, Journal of Construction Engineering and Management, 146(1), p. 04019085. doi: 10.1061/(asce)co.1943-7862.0001736.
11 Yaseen, Z. M. et al. (2020) ‘Prediction of risk delay in construction projects using a hybrid artificial intelligence model’, Sustainability (Switzerland), 12(4), pp. 1–14. doi: 10.3390/su12041514.
12 Health and Safety Executive (2020), ‘Construction statistics in Great Britain’
13 Baker, H., Hallowell, M. R. and Tixier, A. J. P. (2020) ‘AI-based prediction of independent construction safety outcomes from universal attributes’, Automation in Construction, 118(June), p. 103146. doi: 10.1016/j.autcon.2020.103146.

This publication presents the views, thoughts or opinions of the author and not necessarily those of HKA. Whilst we take every care to ensure the accuracy of this information at the time of publication, the content is not intended to deal with all aspects of the subject referred to, should not be relied upon and does not constitute advice of any kind. This publication is protected by copyright © 2021 HKA Global Ltd.