FORECASTING OLEA POLLEN CONCENTRATION BY REGRESSION AND ARTIFICIAL NEURAL NETWORKS

Forecasting airborne pollen concentrations is one of the most studied topics. The most used tools for this problem are regression models. Notwithstanding, few works have used more sophisticated tools based in Neural Networks (NN) models. In this work, we applied some of these models to forecast olive pollen concentrations in the atmosphere of Tirana. After treatment of the Olea pollen data the relation between these and meteorological parameters were studied by multiple regresion. For this a sequence of regression equations was made in order to find the best fitting equation. Each independent variable introduced in the equation explains a new percentage of variance, which is not explained by any other variable. This is the recommended method when the independent variables are intercorrelated, which occurs frequently with meteorological parameters. In this paper we have developed a new forecasting method that applies the ability of NNs to predict the future behaviour of chaotic systems in order to make accurate predictions of the Olea pollen concentration for pre-peak, post-peak and whole periods. This method gave good results for Pearson´s correlation and R-square: the correlation was 0.96, R-square of 0.92 for pre-peak period and 0.82, R-squared 0.76 after the peak correlation, respectively. We used different MPL with three layers and a variable number of hidden nerves. Experimental results show an advantage of the NNs against statistical methods, although there is still room for improvement. Used models gave more satisfactory predictive results, where it was best for the pre-peak, then for post-peak and weak for the whole period.