Air pollution and population morbidity forecasting with artificial neural networks

Abstract

Incidence prediction models for urban population have not yielded consistent or highly accurate results. The complex nature of the interrelationship between "environmental factors and incidence" has many nonlinear associations with outcomes. We explore artificial neural networks (ANNs) to predict the complex interactions between the risk factors of incidence among the urban population. ANN modeling using a standard feed-forward, back-propagation neural network with three layers (i.e., an input layer, a hidden layer, and an output layer) is used to predict the incidences of diseases of children and adults. A receiver-operating characteristic (ROC) analysis is used to assess the model accuracy. We develop a mathematical model taking into account factors of natural, anthropogenic, and social environments. The model effectiveness is proved by computing experiments for the Bratsk industrial centre (Irkutsk region, Russia). Optimal air pollution levels are offered to achieve a background morbidity level among different age groups of the population. The prediction of incidence is most accurate when using the ANN model with several univariate influences on the outcome. An incorporation of some computerized learning systems might improve decision making and outcome prediction.