Comparison of Statistical Methods, Neural Network, and Fuzzy Neural for Particular Matter Prediction: A Case Study in Mashhad, Iran

Abstract

Background: In the recent era, air pollution is a major global concern that affects human health. The emission proportion of air pollutants is increased in many cities of Iran such as Mashhad. Particular matters (i.e. PM2.5) are one of the five air pollutants known to be responsible for polluting the air in Mashhad. Nowadays, fuzzy neural intelligent systems, which are capable of solving nonlinear and complex problems, are widely used in the air pollution problem to determine the amount of the particles and dust in the air. Methods: In the current study, the air quality data consisting of daily average concentrations of air pollutants and the meteorological data including the minimum temperature, precipitation, humidity, wind direction, and daily wind speed recorded by city monitoring stations from 2011 to 2017. The daily average pollutants concentration was used to study the relationship between PM2.5 and the other air pollutants such SO2, O3, NO2, PM10, and CO. SPSS was used for data analysis. Linear regression, multilayer perceptron (MLP) neural network, and fuzzy neural network using MATLAB 2017 software were employed for modeling. Performance of the models was evaluated using root mean square error (RMSE) and coefficient of determination (R2). Results: Based on the obtained results, among the employed models, MLP neural network with R2 = 0.598, RMSE = 0.088, and MSE = 0.0079 was better than linear regression, and the ANFIS model combining particle swarm optimization (PSO) algorithm with R2 = 0.804, RMSE = 0.055, and MSE = 0.0031 had the best performance in the prediction of PM2.5. Conclusions: The ANFIS network correctly fitted more than 80% of total data; given that there were non-linear and complicated models in meteorological systems, this figure can indicate the high strength of ANFIS network through PSO-based combinational training methods for modeling nonlinear data.