An interpretable clustering approach to safety climate analysis: Examining driver group distinctions

Area of Science:

• Occupational Safety and Health
• Transportation Safety
• Data Science and Machine Learning

Background:

• The trucking industry faces significant workplace accidents and fatalities, with large trucks involved in a substantial portion of traffic deaths.
• Safety climate is recognized as vital for accident prevention, yet clustering employees by safety perception remains underexplored.
• Existing studies often lack algorithmic comparison and interpretable methods for understanding factors influencing employee safety perceptions.

Purpose of the Study:

• To introduce an interpretable clustering approach for analyzing truck driver safety climate perceptions.
• To compare five clustering algorithms and propose a novel method for evaluating partial dependence plots (QPDP).
• To enhance the interpretability of clustering results using machine learning techniques like Shapley additive explanations and permutation feature importance.

Main Methods:

• Clustering analysis of safety climate perceptions from over 7,000 American truck drivers using five algorithms.
• Development and application of a quantitative partial dependence plot (QPDP) method for interpretability.
• Utilizing interpretable machine learning measures (Shapley additive explanations, permutation feature importance, QPDP) to explain cluster membership.

Main Results:

• Identified distinct clusters of truck drivers based on their safety climate perceptions.
• Highlighted supervisory care promotion as a critical factor differentiating driver groups.
• Demonstrated the effectiveness of interpretable machine learning in understanding safety climate variations.

Conclusions:

• The study provides an innovative, interpretable clustering approach for safety climate analysis in the trucking industry.
• Findings underscore the importance of supervisory care in shaping driver safety perceptions and suggest targeted interventions.
• Machine learning techniques, particularly cluster analysis, offer valuable tools for advancing scientific knowledge in occupational safety.