Prediction of issuance of emergency evacuation orders for chemical accidents using machine learning algorithm

Abstract

An emergency response to chemical accidents proceeds in the order of prevention, mitigation, preparedness, response, and recovery. One of the methods of response is emergency evacuation orders. To minimize the loss of life, it is important to issue prompt and precise evacuation orders when chemical accidents such as toxic gas emissions occur near populated areas. This paper presents a method and its results for predicting emergency evacuation orders using a naïve Bayes classifier and an artificial neural network. A study was conducted using ATSDR's National Toxic Substance Incidents Program (NTSIP) dataset and The Hazardous Substances Emergency Events Surveillance (HSEES) database by extracting 61,563 of 115,569 accidents that occurred between 1996 and 2014. According to the results of the study, for predicting emergency evacuation orders, Artificial Neural Network prediction had a high level of accuracy when compared to Naïve Bayes Classifier. Based on the Area Under the Curve (AUC) value of the predicted results, the discriminatory power of the model was reliable. These results suggest that using machine learning in the field of chemical process safety can yield meaningful results.