Related Experiment Videos
A new machine learning framework for occupational accidents forecasting with safety inspections integration.
Aho Yapi1, Pierre Latouche2, Arnaud Guillin3
1Laboratoire de Mathématique Blaise Pascal UMR 6620 CNRS, Université Clermont Auvergne, Place Vasarely, 63178, Aubière, France; LYF SAS, 27 rue Raynaud, Clermont-Ferrand, 63000, France.
Journal of Safety Research
|June 15, 2026
Summary
This study introduces a new framework for predicting occupational accidents using safety inspections. The model provides daily and weekly risk scores, enabling proactive safety interventions and resource allocation to prevent incidents.
Area of Science:
- Occupational Safety and Health
- Data Science
- Predictive Analytics
Background:
- Occupational accidents cause significant human harm and financial losses.
- Existing safety programs often fail to proactively anticipate risks.
- Safety inspections are underutilized in predictive risk modeling.
Purpose of the Study:
- To develop a model-agnostic framework for short-term occupational accident forecasting.
- To leverage safety inspection data for dynamic risk assessment.
- To improve proactive risk management in industrial settings.
Main Methods:
- Modeling accident occurrences as binary time series.
- Generating daily predictions aggregated into weekly assessments.
- Employing sliding-window cross-validation for time series data.
- Comparing logistic regression, tree-based models, and neural networks.
Main Results:
- The framework reliably identifies upcoming high-risk periods.
- Robust period-level performance was achieved across tested algorithms.
- Converting safety inspections into binary time series provides actionable risk signals.
Conclusions:
- The methodology generates clear daily and weekly risk scores from inspection data.
- Decision-makers can use scores to prioritize inspections and schedule interventions.
- Proactive interventions based on risk scores can prevent incidents and optimize safety investments.