Rapid response to large-scale emergencies: a neural network model and a decision-tree algorithm

Sánchez Losada JA, Romero Sánchez S, Caamiña García M, Habed Lobos N, Jiménez Carrascosa JF, Touza Garma B, Gil González AM, Sanz Mata P

Objective: The greatest challenge to decision-making during the management of

emergencies with multiple victims is uncertainty in an initially chaotic environment. Our

objective was to develop a predictive model to improve response and early resource

management in the early-phase environment of chaotic uncertainty during large-scale

emergencies.

Methods: A database of information on real incidents with multiple victims in Castile-La

Mancha, Spain, in the last 5 years was used to study the weight of 10 categorical variables

and their effect on the seriousness of the emergencies. A neural network was designed to

learn about these real cases, and a decision tree was generated, to study which of the 2

approaches gave the best results. An important design limitation was that nearly all the

incidents analyzed involved traffic accidents.

Results: The model based on decision-tree analysis gave more information and greater

variability. It proved superior to the neural network, identifying 6 homogeneous groups

according to the following factors: number of initial victims, type of incident, and

environment (P<.05). Conclusions: A predictive model can be based on the considered variables in the interest of improving resource management during a large-scale emergency. However, development based on a larger number of real incidents of different types would be needed before such a model could be applied during real future incidents.

Caamiña García M, Gil González AM, Habed Lobos N, Jiménez Carrascosa JF, Romero Sánchez S, Sanz Mata P, Sánchez Losada JA, Touza Garma B