Data-Driven Machine Learning Approach to Integrate Field Submittals in Project Scheduling

Construction projects are data-rich environments. However, those data are usually captured for site-specific reasons, e.g., the filing and approval of inspection requests, with little regard to how they can be leveraged for improved project management. Typically, scheduling techniques rely on general probability estimates, which do not capture the details of the site processes causing schedule deviations. This paper illustrates how machine learning techniques can mine project data to forecast delay in the midst of the project. The proposed method uses concrete pouring requests as an example of a site data stream and implements a random forest predictive model to forecast the likelihood of acceptance for these requests. Embedded in the proposed approach is an analysis that allows for the addition of probabilistic time delays associated with the forecast of rejected requests. The methodology was tested on a real-world case study, allowing for the comparison between a project duration estimate based on critical path method (CPM) with static buffers and a project duration obtained using the proposed method. The results show a difference of 10% between the two durations. The paper shows how using data streams from a construction site with machine learning techniques can enhance project duration estimates in execution.

Automated summary

This paper explains how machine learning techniques can be used to mine construction project data and forecast project delays, using concrete pouring requests as an example and implementing a random forest predictive model to predict the likelihood of acceptance for these requests. The proposed method includes an analysis for adding probabilistic time delays associated with rejected requests, and was tested on a real-world case study showing a 10% difference between project duration estimates based on critical path method (CPM) with static buffers and those obtained using the proposed method.