Citation

Zhu Y, Liao H, Huang D. Accid. Anal. Prev. 2023; 191: e107224.

Copyright

(Copyright © 2023, Elsevier Publishing)

DOI

10.1016/j.aap.2023.107224

PMID

37506406

Abstract

Incident investigation reports provide information on defects related to the system safety and indications for improvements. Currently, the analysis of these reports relies heavily on expert' experience. The foreseeable work-load and lack of understanding about the importance of near misses have created a situation where severe accidents are rigorously investigated, and minor incidents are often omitted. Consequently, incident reports have not been fully analyzed to provide sufficient solutions. The aim of this research is to propose a framework that uses text mining and multilevel association rules to efficiently structure Chinese incident reports and identify important incident patterns, providing an analysis of trends, rectification strategies, and guidance for safety management. A case study of a construction company in China was conducted using two years of incident data dated 2018-2019, including accidents and near misses. To identify incident elements, a pattern extraction workflow involving TextRank, and domain pertinence was devised based on the linguistic and writing styles of Chinese reports. A concept hierarchy was applied to determine the taxonomic relationships within the risk factors. Multilevel association rule mining was adopted and proven to deliver more comprehensive pattern indications. Comparative and cross-analysis of patterns in different time periods revealed the severity and temporal features of incidents as well as the effectiveness of preventive and precautionary measures. The results also highlight the importance of learning from near miss events. Decision makers can formulate countermeasures and management policies based on these results to improve safety performance.

Language: en

Keywords

Near miss; Text mining; Concept hierarchy; Incident analysis; Multilevel association rule; The effectiveness of safety measures