Factors influencing unsafe behaviors: A supervised learning approach

Abstract

Despite its potential, the use of machine learning in safety studies had been limited. Considering machine learning's advantage in predictive accuracy, this study used a supervised learning approach to evaluate the relative importance of different cognitive factors within the Theory of Reasoned Action (TRA) in influencing safety behavior. Data were collected from 80 workers in a tunnel construction project using a TRA-based questionnaire. At the same time, behavior-based safety (BBS) observation data, % unsafe behavior, was collected. Subsequently, with the TRA cognitive factors as the input attributes, six widely-used machine learning algorithms and logistic regression were used to develop models to predict % unsafe behavior. The receiver operating characteristic (ROC) curves show that decision tree provides the best prediction. It was found that intention and social norms have the biggest influence on whether a worker was observed to work safely or not. Thus, managers aiming to improve safety behaviors need to pay specific attention to social norms in the worksite. The study also showed that a TRA survey can be used to extend a BBS to facilitate more effective interventions. Lastly, the study showed that machine learning algorithms provide an alternative approach for analyzing the relationship between the cognitive factors and behavioral data.

Keywords: Construction safety; Machine learning; Safety behavior; Theory of reasoned action; Working at height.