Text Mining-Based Association Rule Mining for Incident Analysis: A Case Study of a Steel Plant in India

Sarkar, Sobhan; Vinay, Sammangi; Djeddi, Chawki; Maiti, J.

Although a large amount of accident data in terms of categorical attributes and free texts are available across large enterprises involving high-risk operations, the methodology for analyzing such mixed data is still under development. The present study proposed a new methodological approach to extract useful inherent patterns or rules for accident causation using association rule mining (ARM) of both incident narratives (unstructured texts) and categorical data. Incidents data from an integrated steel plant for a period of four years (2010–2013) are used for model building and analysis. In the first phase, the text mining approach is employed to find out the basic events that could lead to the occurrences of faults or incident events. In the second phase, text-based ARM has been used to extract the useful rules from unstructured texts as well as structured categorical attributes. A total of 23 best item-set rules are extracted. The findings help the management of the plant to augment the cause and effect analysis of accident occurrences as well as quantifying the effects of the causes, which can also be automated to minimize the human involvement.