Text-clustering based deep neural network for prediction of occupational accident risk: A case study

Predicting occupational accident risk using both structured and unstructured (text) data is broadly an unexplored area of research. Unstructured texts, i.e., incident narratives often remain unutilized or under-utilized. Besides the explicit attributes present in the dataset, there exist a large number of hidden attributes in different forms, which are hardly explored by the traditional machine learning algorithms. Therefore, we propose a methodology that utilizes both text-based clustering, namely Expectation Maximization (EM) algorithm for unstructured text analysis and deep neural network (DNN) for prediction of accident risk using the accident data collected from a steel plant in India. EM-based DNN shows the maximum accuracy equal to 83.59% in the prediction of risk while compared to other algorithms, namely single DNN, support vector machine, and random forest. In addition, it is also explored that the use of text data enhances the prediction accuracy in accident analysis.