A combined approach for the analysis of large occupational accident databases to support accident-prevention decision making

Occupational accidents are commonly collected in large databases by National Workers Compensation Authorities and companies’ safety and prevention teams. The analysis of the data can be difficult because the database elements are characterized by many parameters, which are not of a numerical nature. Data mining techniques could represent an efficient tool for the identification of useful information in large databases. In 2011, a two-level clustering method, made of SOM and numerical clustering, obtained positive results in identifying critical accident dynamics. The present research proceeds from that initial methodology. A sensitivity analysis of the coupled clustering method was carried out. Some improvements have been designed, and an enhanced methodology has been applied to the original case study data set, for validation purposes. This method represents an efficient tool for the analyst that has to deal with the occupational accidents data, thanks to its capability of grouping and visualizing data in a readable and exportable outcome. The information acquired by this method can help analysts to better address the measures to be adopted in a work environment, in order to prevent occupational accidents.