Predictive model for incident occurrences in steel plant in India

Steel industry is considered to be an economic sector with higher number of accidents. Workers in this industry are exposed to a wide variety of hazards during working hours. Thus, the database maintained in industry varies in terms of the types of data indicating nature of accidents, causes of accidents, date and time-stamp etc. The objective of this study is to give predictive solution of accident occurrences in a steel industry based on free-text data or narratives logged in the database from previous incidents report. Utilizing the database comprised of 9488 observations of occupational accidents during the year 2010 to 2013 in an integrated steel manufacturing plant in India, text mining technique has been used, and its outputs have been fed into three classifiers namely Support Vector Machines, Random Forest, and Maximum Entropy which are tested for the evaluation of best fit model that could predict the injury occurrences in workplace. Furthermore, with the help of the same three classifiers, the causes of injuries are also predicted. Maximum Entropy and Random Forest classifiers are found out to be the best of all other classifiers in terms of higher area under curve (AUC) value in binary and multi-class prediction model, respectively.