Towards a Deeper Understanding of the Cost Performance of Large-scale Road Projects

Abstract

The issue of cost overruns in large-scale infrastructure projects is a major concern for many countries, including Norway. Despite efforts to manage project costs, cost inaccuracy remains a persistent problem in road construction projects, resulting in significant financial losses, delays, and negative social and environmental impacts. Therefore, it is essential to identify the factors contributing to cost inaccuracy and develop effective cost management strategies to ensure the success of these projects. This thesis aims to address this problem by exploring the critical importance of cost performance evaluation in the Norwegian road projects’ governance framework. Two broad research objectives have guided the research process for this thesis. The first research objective (RO1) was to further understand the concept of cost overrun and the influential parameters affecting cost inaccuracies in Norway. The second research objective (RO2) was to predict the cost performance of largescale road (LSR) projects using past project data (a data set of 52 LSR projects). To achieve these objectives, the research employed a variety of methods, including surveys, case studies, and data analysis using advanced machine learning techniques such as artificial neural networks (ANNs) and generative adversarial networks (GANs). The research findings revealed the three main reasons for cost overruns in road projects during the construction phase in Norway: scope changes, market conditions, and unforeseen ground conditions. These factors can significantly impact the project’s budget, and it is essential to consider them during the planning phase to minimize the risk of cost overruns. The research also found that it is necessary to consider not only the construction phase but also the planning phase, as many factors that can cause cost overruns often originate from decisions made during the planning phase. To minimize the risk of cost overruns due to scope changes and market conditions, the research suggested several strategies, including well-defined project scope, open communication, collaboration, thorough risk analysis, monitoring market conditions, local sourcing of materials and equipment, and collaboration with suppliers, subcontractors, and stakeholders. It is also crucial to conduct thorough geological and geotechnical investigations before starting construction and have a contingency plan in place to minimize the risk of unforeseen ground conditions causing cost overruns. The research also suggested revising the calculation method used in Norway to account for cost increases in the planning period and/or budget increases during the construction phase. The current method may not be a reflection of the actual cost performance of a project, and including ex-post evaluations in project governance systems to assess tactical and long-term strategic success could provide a more accurate assessment of project success. To improve on traditional cost estimation models which may be inadequate for capturing road projects’ complex and dynamic nature, machine learning (ML) algorithms were explored. ML algorithms were seen to improve accuracy in analysis and feature selection, which helps identify important factors contributing to cost overruns while eliminating irrelevant or redundant variables. This thesis also shows the potential for deep learning methods to address small sample size challenges and the advantages of using ANNs over traditional regression models in predicting the cost performance of LSR projects. The results showed that the ANNs significantly outperformed traditional regression models in predicting cost performance. The study employed data generation techniques, specifically Conditional Tabular GAN (CTGAN), to overcome the challenge of small sample size in cost performance prediction. This approach facilitated the training and evaluation of three classifiers, yielding impressive results in terms of high accuracy and F1 scores across all classifiers. Overall, the use of these new and more sophisticated algorithms showed promise and could be beneficial in overcoming the challenges of cost estimation and management in the Norwegian road construction industry. In conclusion, this thesis provides critical insights into the factors contributing to cost inaccuracy in Norwegian road projects and offers promising avenues for improving cost performance evaluation using advanced machine learning techniques. The findings of the research can inform the development of effective cost-management strategies and contribute to the successful delivery of large-scale infrastructure projects in Norway and other countries facing similar challenges. However, further research is needed to explore and refine these methods and strategies to ensure their effectiveness and applicability in different contexts.