Comparing Statistical and Neural Network Approaches for Urban Air Pollution Time Series Analysis

D. Dunea, M. Oprea, and E. Lungu (Romania)


Urban air quality, feed-forward neural network, back propagation, statistical models, time series, ARIMA


The paper presents an analysis of the performances obtained by using an artificial neural networks model and several statistical models for urban air quality forecasting. The time series of monthly averages concentrations (Sedimentable Dusts, Total Suspended Particulates, Nitrogen Dioxide, and Sulfur Dioxide emissions) recorded between 1995 and 2006 in the urban area of Târgovişte were used as inputs in these models. The original measured pollutant data were statistically analyzed in time series including monthly and seasonal patterns using the auto-regressive integrated moving average (ARIMA) method, linear trend, simple moving average of three terms and simple exponential smoothing. The performance evaluations of the adopted statistical models were carried out and discussed according to the root mean square error (RMSE) estimations and several tests. Due to their generalization capacity, ANN has been proposed in this work as model for time series forecasting, because ANN provided better air quality forecasts than the Box-Jenkins ARIMA method. Advantages of neural computing techniques over conventional statistical approaches relied on faster computation, learning ability and noise rejection. The forecasted values are satisfactory and the presented technique promises perspectives for air quality forecasting.

Important Links:

Go Back