Predicting the length of a post-accident absence in construction with boosted decision trees

Abstract

Work safety control and analysis of accidents during construction performance are one of the most important issues of construction management. The paper focuses on post-accident absence as an element of occupational safety management. Somehow, the length of the post-accident absence can be treated as an indicator of building performance safety. The paper attempts to answer the question of whether it is possible to use boosted classifier ensembles to predict the post-accident absence length using a small set of historical observations, and which classification algorithm is the most promising to solve the prediction problem. It also proves that there is a dependence between the length of the post-accident absence and the cause of the accident or working conditions The choice of boosted algorithms is not accidental. Thanks to the use of aggregation methods it is possible to build classifiers that predict precisely and do not require any initial data treatment, which simplifies the prediction process significantly. The model of the prediction problem has been clarified. To identify the most promising classifier ensemble the prediction accuracy measures of selected classification algorithms were analyzed. The data used to build models was gathered on national (Polish) construction sites.

Keywords

• classifier ensembles
• post-accident absence
• boosting