A Multi-Output Regression Approach to Predicting Gender-Centric Workplace Fatal Accidents

Abstract

The prevalence of accidents that raise substantial concerns, due to their potential to cause severe injuries, loss of life, and long-term health implications for workers across various industries highlights the urgent need for effective accident prevention measures. It is imperative to analyze accident characteristics and different features leading to accidents, hence, this study investigates the involvement of men and women in different industry accidents and the contributing factors. The number of men and women involved in accidents is examined through the application of linear regression and tree-based methods, including Random Forest, Extra Trees, and XGBoost Regressors, within the context of constrained multi-output regression based on the data from the year 1992 to 2015. A model that predicts gender-related accidents for the year 2016 to 2021 has been developed using historical data and its accuracy is compared to the actual data from that time. The error calculation has been determined using the Mean Average Percentage Error (MAPE) method, the results showing an error of 2.57%, 9.73%, and 2.86% in the prediction of the number of accidents for males, females, and total workers, respectively. Moreover, the feature importance analysis is conducted to identify the key factors influencing the number of accidents for both genders. The analysis reveals that the number of wage and salary workers have the highest importance for both genders, followed by the number of working-age and under-age workers for accident prediction of male and female workers, respectively. This study aims to gain insights into the underlying factors contributing to accident occurrence and enhance the understanding of accident involvement among men and women. The study results can contribute to the development of effective strategies for accident mitigation, worker protection, and the promotion of safer working environments by predicting accidents using machine learning models.

Keywords

• workplace accident
• accident prediction
• occupational safety
• gender-based accident
• mean average percentage error approach
• feature importance