A0563
Title: Forecasting cardiorespiratory hospitalizations from air pollution levels through artificial neural networks
Authors: Andrea Bucci - universita degli studi g d annunzio di chieti pescara (Italy) [presenting]
Luigi Ippoliti - University G.d'Annunzio Chieti-Pescara (Italy)
Pasquale Valentini - University G. d Annunzio of Chieti-Pescara (Italy)
Abstract: Air pollution is one of the most threatening risk factors for human health conditions. In fact, the epidemiological literature has widely proved that exposure to high levels of air pollution is associated with an increase in mortality and cardiorespiratory hospitalizations. Understanding how and if peaks in air pollution levels are capable of anticipating hospitalizations or death counts can be of great interest for policymakers to define public health strategies. In this context, the aim is to investigate the predictability of hospitalizations by cardiorespiratory diseases in Italian Provinces through the levels of ambient air pollution, such as nitrogen dioxide, sulfur dioxide and particulate matter. Since such a relationship is neither linear nor easy to be predicted, we propose to use neural networks to obtain the predictions of cardiorespiratory hospitalizations. In fact, neural networks can approximate any linear or nonlinear relationship and it has been already shown how they provide accurate time series predictions. Furthermore, recent extensions of traditional neural networks also allow accounting for spatial effects, which is a well-known characteristic of environmental phenomena. We compare their predictive accuracy with traditional neural networks and traditional spatio-temporal models.
