(54bo) Faster and More Accurate Responses for Chemical Accidents Using Machine Learning Technique

Today, chemical accidents are great concern of worldwide owing to recent chemical disasters. In order to minimize loss of life, it is crucial to develop a rapid and correct evacuation order when chemical accidents such as toxic gas releases occur near populated areas.

In this paper, a method and results are presented to perform the prediction of emergency evacuation orders by utilizing reliable and good-quality data from chemical incidents. The Hazardous Substances Emergency Events Surveillance (HSEES) system has collected and analyzed information about acute releases of hazardous substances that result in a public health action, such as an evacuation in the U.S. Data were analyzed using the Naïve Bayes classifier(NBC) and Back propagation algorithm. This study was performed with the aid of Rapidminer 7.5 software, a big-data analysis tool. The analysis included over 61,563 incidents cases among which approximately 10% are the cases of actual evacuation orders. The study was also carried out to verify overall percentage agreement, sensitivity, specificity and Area Under Curve(AUC) values. As results, it is figured out that effectiveness using back propagation algorithm is better than that using NBC. Furthermore victims’ info have been used for evaluating algorithms. The results from this study would help future evacuation orders in case of chemical disasters for better accuracy.