Feature Weights for Contractor Safety Performance Assessment: A Comparative Study of Expert-Based and Analytics-Based Approaches

Abstract

Monitoring contractor safety performance by evaluating various safety management elements is critical for construction site safety. However, current expert-based approaches to determine the weights of different safety management elements can be biased, time-consuming and lack rigorous validation. Hence, using a dataset of 461 data points and 12 features, we explored the use of supervised learning, cluster-then-predict and two-level variable weighting K-Means (TWKM) and compared them with the Delphi method. Supervised learning and TWKM are promising, showing improvements of 20% and 14% against the benchmark, respectively. This study also highlights the importance of constant data monitoring to address covariate shifts across different construction stages and projects, which can reduce the validity of the weights. Key contributions of this study include the analytics-based approaches to develop weights for measuring contractors’ safety performance, and proposed strategies to manage impact of covariate shifts on accuracy of feature weights.

Keywords: Construction safety, safety management system, safety performance evaluation, machine learning, Feature weights