Citation

Li X, Zhu R, Ye H, Jiang C, Benslimane A. Safety Sci. 2021; 140: 105315.

Copyright

(Copyright © 2021, Elsevier Publishing)

DOI

10.1016/j.ssci.2021.105315

PMID

unavailable

Abstract

In recent years, many scholars have used data mining algorithms to discover the laws related to the prevention of occupational injuries in the construction industry. Using accident injury records to model occupational risk is a promising direction (for example, predict injury risk from accident frequency and severity). However, the records of specific accident injury data are relatively limited, bringing great difficulties for people to obtain risk knowledge and establish an effective accident consequence prediction model. This article proposes a meta-learning framework called MetaInjury, which can help safety managers share risk knowledge and predict the risk of work-related injuries in various construction industry accidents. When facing small sample data of a new accident type, we first calculate the document vector of the accident description and compare its center vector with vectors in the Meta-knowledge database to find the type of accident most similar. Then, we correspond the meta-features with the best machine learning algorithms on the data set to implement the recommendation of accident prediction algorithms. Through finding the most similar cases and the recommended algorithm, important accident risk factors and accident assessment rules can be shared by safety managers to realize effective risk management. Moreover, specific small sample accident consequences can be predicted by the recommended algorithm. Finally, we verify the method's effectiveness in four different small sample accident data. The results show that the MetaInjury framework can provide theoretical support for preventing small sample accidents and injury reduction in the construction industry.

Language: en

Keywords

Decision support; Machine learning; Occupational injury; Prediction; Risk analysis; Small sample learning