A machine learning and data analytics approach for predicting evacuation and identifying contributing factors during hazardous materials incidents on railways

Abstract

An emergency evacuation order might be issued in response to a railway incident involving hazardous materials (hazmat), such as the February 2023 derailment at Palestine, Ohio. Due to the potential complexity of railway incidents involving hazmat, making an accurate and timely decision regarding evacuation orders can be very challenging. An appropriate framework is required to predict the need for evacuation after railway incidents and identify the contributing factors. This study aims to develop such a framework by incorporating machine learning techniques. First, various supervised machine learning models are implemented to analyze the effect of different factors on the prediction of evacuation immediately after railway incidents. Based on the factors considered in this study, the most accurate model is identified for predicting evacuation. This model is also used to identify the most significant factors affecting evacuation. The rules leading to the evacuation are then recognized along with the underlying causes of the evacuation. Second, natural language processing and co-occurrence network analysis are employed to analyze brief descriptions of railway incidents that resulted in the need to evacuate. This allows us to construct a network of causes and contributing factors to the evacuations and demonstrate a causal relationship between them. This study provides prediction models and conclusions that can be used to reduce the risks associated with the railway transportation of hazmat and enhance the effectiveness of safety intervention measures.