A systematic approach to develop safety-related undesired event databases for Machine Learning analyses: Application to confined space incidents

Abstract

In Occupational Safety and Health (OSH) and operational safety, confined spaces are high-risk working areas, where frequent serious and fatal incidents occur. However, there is a limited use of data-driven approaches based on Machine Learning (ML) techniques for learning from such incidents. In this context, our study proposes a systematic approach to develop a structured database suitable for ML-based analyses, considering various data sources with unstructured or semi-structured data. The approach was applied to 1346 incidents that happened in confined spaces mainly containing chemicals and located in manufacturing and process industries. This allowed building models predicting the fatal or non-fatal consequences and direct causes of incidents. The recognition of the approach potentialities and difficulties stimulated the development of an improved version of it to be adapted to different safety-related undesired events. The generalised systematic approach supports safety analysts during the process of learning from undesired events to maximise the lessons learned.