A machine learning-based framework for delay risk mitigation in building projects

Abstract

The construction industry is lagging behind the service and manufacturing industries in terms of efficiency and productivity; though the industry continues to boom, as demonstrated in the rapid growth of tall buildings in urban centres across the globe. The rise of tall buildings is partly in response to the need to create more urban space for an impeding global population explosion and urbanization; however, this building typology is notorious for being delayed and uncompleted. The research domain is saturated with numerous studies on construction delays across continents and project types. These studies only make modest contributions in dealing with the inherent problem. An inadequate effort has been channelled towards the development of prescriptive tools with the potential to mitigate construction delays. The desired solution would employ innovative methods to arrive at problem-solving strategies for the ultimate purpose of delay mitigation. Furthermore, the current mantra of the construction industry is to embrace the fourth industrial revolution (IR 4.0), and leverage the capabilities of digital technologies such as artificial intelligence and machine learning. Thus, the aim of this study was to develop a delay mitigation framework based on the application of machine learning, with a focus on tall building projects. The proposed framework is dependent on three key areas of delay mitigation, including -reliable cost estimates?, -reliable duration estimates?, and -delay risk assessment?. This was achieved in two phases of data collection and model development. The first phase identified the causes of delay in the global construction industry, and subsequently determined the delay risk factors in tall building projects. Likewise, historical data on completed tall building projects featuring the total cost and duration of the project was obtained. In the second phase, machine learning models were developed based on Multi Linear Regression Analysis (MLRA), Artificial Neural Networks (ANN), K Nearest Neighbours (KNN) and Support Vector Machines (SVM). This stage also involved combining models to develop ensemble models/multi classifier systems to investigate the possibility for improved predictive performance. The developed models were evaluated by standard performance metrics used in machine learning for classification and regression problems (i.e. Classification Accuracy, Correlation Coefficient (CC), Mean Absolute Percentage Error (MAPE), and Root Mean Squared Error (RMSE)). The performance of the selected model for the duration was characterised by a CC of 0.69, MAPE of 0.18 and RMSE of 301.76, while for cost by a CC of 0.81, MAPE of 0.89 and RMSE of 6.09, and for delay risk a classification accuracy of 93.75% was achieved. The final visualization of the delay mitigation framework was conveyed with the most prevalent analytics model: the Cross-Industry Standard Process for Data Mining (CRISP-DM). Finally, the proposed framework was reviewed and validated by industry professionals. The significance of the proposed framework can be seen in its potential as a decision-making tool for proactive delay risk mitigation at the planning stage of tall building projects. Although the development of delay mitigation framework concentrated on tall building projects, a similar approach can also be extended to other types of construction.