Application of hybrid clustering technique for pattern extraction of accident at work: A case study of a steel industry

Abstract

The phenomenon of occupational accidents is a serious concern of any industry. There are various factors present which collectively interplay behind the occurrence of accidents. For the purpose of prevention of accidents, data are collected and analyzed. But, the higher dimensional data makes the task often difficult, even visualization of data becomes obscure. In addition, unstructured accident text data are most of the time found to be unutilized or under-utilized. In order to handle the issues, the present work employs topic modeling technique to handle the large amount of accident narratives. Self-organizing map (SOM), used for better visualization of the data, is used with other clustering algorithms like k-means, hierarchical clustering for extraction of accident patterns. SOM-based k-means outperforms other algorithms with 84.52% accuracy. In addition, six optimal number of clusters are found out from the analysis. Slip-trip-fall, and material handling are found to be the significant determinants towards the accident causation. The proposed approach compared with other algorithms is found to be effective and can be applied in other industries.