Factors influencing unsafe behaviors: A supervised learning approach.

Area of Science:

• Occupational Health and Safety
• Cognitive Psychology
• Machine Learning Applications

Background:

• Limited use of machine learning in safety studies despite its predictive potential.
• Behavior-based safety (BBS) often relies on direct observation, which can be extended with cognitive insights.
• The Theory of Reasoned Action (TRA) provides a framework for understanding behavioral determinants.

Purpose of the Study:

• To evaluate the predictive accuracy of machine learning algorithms for safety behavior.
• To determine the relative importance of TRA cognitive factors in influencing safety behavior.
• To explore the integration of TRA surveys with BBS for enhanced safety interventions.

Main Methods:

• Collected TRA questionnaire data and BBS observation data (% unsafe behavior) from 80 construction workers.
• Developed predictive models using six machine learning algorithms and logistic regression with TRA factors as input.
• Utilized Receiver Operating Characteristic (ROC) curves to assess model performance.

Main Results:

• The decision tree algorithm demonstrated the best predictive accuracy for unsafe behavior.
• Worker intention and social norms were identified as the most significant cognitive factors influencing safety behavior.
• Machine learning models effectively analyzed the relationship between cognitive factors and observed behavior.

Conclusions:

• Machine learning offers a valuable alternative for analyzing cognitive influences on safety behavior.
• Safety interventions should prioritize addressing social norms and enhancing worker intention.
• TRA surveys can complement BBS, leading to more effective safety management strategies.