Learning from accidents: machine learning prototype development based on the CRISP-DM business understanding
Paper in proceeding, 2021

Occupational accidents continue to be an unresolved problem in the Swedish construction industry, despite a whole range of routines, campaigns, education, management appraisals, authorities’ enforcement, networks, and research in place. While registered accidents are less frequent, there is a widespread willingness to strive for better performance. A potential solution is to apply more robust data analytics to the large company occupational accident registers, complementing existing regular analysis. Machine learning (ML) can provide a promising solution for strengthening data analysis, and international prototypes of such systems are emerging. However, there is a need to appreciate local and corporate concerns, and the ML development method “Cross Industry Standard Process Development Method” (CRISP-DM) appears to offer just that. This paper aims to analyse experiences and challenges in using the first phase of CRISP-DM, i.e., “business understanding”. The sociomaterial approach serves as the framework of understanding and is supplemented with accident research and ML development concepts. Methodologically, the paper draws on an ongoing research project to develop a ML prototype for occupational accident analysis. It quickly surfaced that CRISP-DM’s “business understanding”, while asking relevant questions in the company context (such as the goal for the model and the relative application), was too general to provide developmental guidelines. We, therefore, shifted from a top-down to a bottom-up approach, where knowledge on accident registration procedures and registered accidents became the starting point for iterative prototype development. Also, early challenges were to understand the registered data extracted from standard software with limited transparency, and tackle register entries of different quality. Apart from CRISPDM’s slightly idealistic approach to a company context, it is important to appreciate the classical decoupling between top management and (bottom) project levels in Swedish contractor companies.

accident register

accidents

machine learning

construction

CRISP-DM

Sweden

Author

May Shayboun

Chalmers, Architecture and Civil Engineering, Building Design

Christian Koch

Aarhus University

Dimosthenis Kifokeris

Chalmers, Architecture and Civil Engineering, Building Design

Proceedings of the Joint CIB W099 & W123 Annual International Conference 2021: Good health, Changes & innovations for improved wellbeing in construction

43-53 51
978-1-91418-801-5 (ISBN)

Joint CIB W099 & W123 Annual International Conference 2021: Good health, Changes & innovations for improved wellbeing in construction
Online, ,

A new generation of preventive measures of occupational accidents with machine learning

Development Fund of the Swedish Construction Industry (SBUF) (13670), 2019-06-01 -- 2021-06-01.

Subject Categories

Construction Management

Environmental Health and Occupational Health

More information

Latest update

1/19/2022