Artificial neural networks: an application in the study of air pollution and its adverse health effects

Emília Matos do NASCIMENTO[1]

Basilio de Bragança PEREIRA1,[2]

José Manoel de SEIXAS[3]

§    ABSTRACT: There is a great need to assess the association between weather and air pollution with mortality or hospital admissions due to respiratory diseases. This paper proposes neural networks as alternative methodology to evaluate that association. The data refer to the number of hospitalizations in the city of Paris due to infant bronchiolitis, between 1997 and 2000. The neural models were evaluated for data description and to measure their capacity of generalization. The best results were obtained through data pre-processing, with removal of cycles and use of a moving average filter. A relevance study of the explanatory variables was also carried out. The results were consistent to those found through generalized additive models pointing out the particulate matter (PM10) as the main responsible for the number of hospital admissions.

§    KEYWORDS: Artificial neural network, air pollution; respiratory diseases.

 



[1] Programa de Engenharia de Produção, COPPE/UFRJ, Caixa Postal: 68507, CEP. 21941-972 - Rio de Janeiro - RJ – Brasil. E‑mail: emilia@pep.ufrj.br / basilio@hucff.ufrj.br

[2] Faculdade de Medicina, HUCFF/UFRJ, Rio de Janeiro, RJ,  Brasil

[3] Laboratório de Processamento de Sinais, Programa de Engenharia Elétrica, COPPE/Poli-UFRJ, Caixa Postal: 68504, CEP. 21941-972, Rio de Janeiro, RJ,  Brasil. E‑mail: seixas@lps.ufrj.br