Text-document clustering-based cause and effect analysis methodology for steel plant incident data.

Area of Science:

• Industrial Safety and Health
• Data Mining and Text Analytics
• Occupational Risk Management

Background:

• Traditional cause-effect analysis relies on expert knowledge, potentially missing comprehensive incident root causes.
• Incident reports from workers offer a rich, underutilized data source for identifying causal factors.
• Existing methods may not fully capture the nuances and hidden trends within incident descriptions.

Purpose of the Study:

• To develop a novel text clustering-based methodology for cause and effect analysis of incident data.
• To identify root causes and sub-causes of workplace incidents using natural language processing techniques.
• To explore the utility of text clustering for uncovering hidden factors and anomalies in incident reporting.

Main Methods:

• Collected incident report data from an integrated steel plant.
• Applied singular value decomposition (SVD) for dimensionality reduction of text data.
• Utilized the expectation-maximization (EM) algorithm for text-document clustering.

Main Results:

• Text-document clustering proved to be a feasible method for analyzing incident report descriptions.
• The methodology successfully identified hidden factors and trends contributing to workplace incidents.
• Anomalies and inconsistencies within the incident reporting process were detected.

Conclusions:

• Text clustering offers a powerful, data-driven approach to complement traditional cause and effect analysis.
• This methodology enhances the understanding of incident causation by leveraging worker-generated text data.
• The findings suggest improvements in incident data collection and analysis for enhanced workplace safety.