Constructability of Districts: Capabilities of Productivity and Logistics Big Data for Machine Learning Prediction

Big data, reflecting both qualitative information and quantitative material, can be used within the construction management processes of complex and large-scale building activities, such as the development of whole districts in urban areas. Such big data is probably largely focused on transport routes, productivity and site logistics portfolios. However, despite the capabilities offered by construction informatics, such data has scarcely been utilized systematically and in its full capacity for descriptive and predictive purposes. Such a systematic data utilization process can be framed through the lens of the novel construction management concept of district constructability, namely the extension of constructability into the collective level of entire districts. Constructability is here understood as the optimal use of construction knowledge and experience in planning, design, procurement, and field operations, to achieve the project objectives of time, cost and quality, and omit the gap between the as-designed and as-built project states. District constructability moves from individual projects to an overall metric for the facilitation of construction knowledge and experience implementation when undertaking large-scale construction activities (e.g. the erection of numerous buildings) for the development of entire districts; thus, it can be realized, among others, through the achievement of optimal construction productivity rates and smooth logistics operations. To combine all the aforementioned, and simultaneously fully and meaningfully exploit the capabilities that construction productivity and logistics big data may present for the assessment of district constructability, data mining can be utilized, namely the set of processes that computationally discover and “comprehend” patterns in datasets. More particularly, machine learning, here defined as the exploration of algorithms that enable computing systems to “learn” and make data-driven predictions by building a model from a sample dataset and without being explicitly programmed, can be at the methodological forefront of fully exploiting all data found in transport routes, buffer facilities, productivity rates and logistics portfolios. In this paper, the capabilities of the information structures found in the data for developing machine learning models predicting the district constructability in new large-scale urbanization activities, are examined.