gene expression programming algorithm

As one of the hazardous pollutants, ozone (O3), has significant adverse effects on urban dwellers' health. Predicting the concentration of ozone in the air can be used to control and prevent unpleasant effects. In this paper, an attempt was made to find out two empirical relationships incorporating multiple linear regression (MLR) and gene expression programming (GEP) to predict the ozone concentration in the vicinity of Zrenjanin, Serbia. For this purpose, 1564 data sets were collected, each containing 18 input parameters such as concentrations of air pollutants (SO2, CO, H2S, NO, NO2, NOx, PM10, benzene, toluene, m- and p-xylene, o-xylene, ethylbenzene) and meteorological conditions (wind direction, wind speed, air pressure, air temperature, solar radiation, and relative humidity (RH)). In contrast, the output parameter was ozone concentrate. The correlation coefficient and root mean squared error for the MLR were 0.61 and 21.28, respectively, while the values for the GEP were 0.85 and 13.52, respectively. Also, to evaluate these two methods' validity, a feed-forward artificial neural network (ANN) with an 18-10-5-1 structure has been used to predict the ozone concentration. The correlation coefficient and root mean squared error for the ANN were 0.78 and 16.07, respectively. Comparisons of these parameters revealed that the proposed model based on the GEP is more reliable and more reasonable for predicting the ozone concentrate. Also, the sensitivity analysis of the input parameters indicated that the air temperature has the most significant influence on ozone concentration variations.