Deep neural network-based probabilistic classifier of occupational accident types on a construction site in Korea

Abstract

The number of accidents in the Korean construction industry has been increasing rapidly, reaching about 25,000 every year. Although strong and binding laws and systems have been implemented to reduce accidents in the construction industry, the frequency of accidents is still higher than that of other industries. While existing studies have provided models for predicting occupational accidents, there are limitations in predicting individual workers’ occupational accidents specific to site conditions and worker tasks. This study proposed a deep neural network model that can classify and predict core accident types for workers in construction sites, considering the characteristics of the site and workers based on 70,204 case data from the Korea Occupational Safety and Health Agency. The result of this study would contribute to improve safety in construction sites by being used for safety accident prevention training customized for workers at construction sites. Additionally, the model can be expanded to develop a real-time construction site occupational accident monitoring and early warning system by integrating ICT-based safety management technologies such as wearable devices and CCTV.

Keywords

• occupational accidents
• deep neural network
• customized probabilistic prediction
• classifier of accident types