On Effective Text Classification for Supporting Job Hazard Analysis

Abstract

The construction industry is known to be dangerous and is featured by fatal occupational hazards. Job Hazard Analysis (JHA) is a common approach to mitigate and control these occupational hazards. It analyzes major tasks in a construction activity, identifies all the potential task related hazards, and suggests means to reduce or avoid each potential hazard. Because every project is unique, an effective JHA must consider the conditions specific to the project instead of reusing previous JHAs directly. To this end, JHA requires the participation of experienced construction practitioners and becomes a time-consuming and brain-draining task. While expert involvement is necessary during JHA development, the authors would like to explore the possibility of leveraging existing construction safety resources in order to reduce the required human efforts. This paper presents an approach based on text classification to support the automation of JHA. It uses the CPWR construction solutions database as an example to demonstrate how text classification can be applied to match the database documents with predefined safety violation scenarios and identify potentially useful safety approaches. This paper also discusses how different strategies were tested to optimize the effectiveness of the proposed text classification approach. The results indicate that although some of the classification strategies cannot make obvious progress on the effectiveness, the effectiveness of the text classification without optimization is good enough to support JHA.