Analysis of Neural Network models for Prediction of Air Pollution Index

Abstract

Environmental analysis focusing on air pollution becomes an emerging concern to the environmental analyst. This includes gathering and analysis of data to predict the model for preventive actions. Air pollution Index (API) is a mechanism that provides better understanding about air pollution. The API increases due to industrialization and urbanization. It influences a natural danger at a quick rate. Therefore, the air quality forecast became a necessary measure for both environmental analyst and for the society.  In the existing research, various neural network techniques are used to forecast air pollution; however comparative analysis of these techniques is often required to have better understanding of their processing time for multiple datasets. In this paper, the selected neural network models such as ANN, DLR, LSVM, DBN, ANNGA, BPNMLP and FTS are compared for the same dataset considering their accuracy and processing time. These models are tested using MATLAB neural network simulation toolbox and their performance are discussed considering the air pollutants such as Ozone (O3), Nitrogen Oxides (NO2), Carbon Monoxide (CO), Sulfur Dioxide (SO2) and Particulate Matter (PM10 and PM2.5). This dataset represents Telangana state pollution control board, Hyderabad. Mean absolute predicator error (MAPSE) and Root mean square error (RMSE) have been used as an evaluation criterion for the comparison of ANN models. It produces the effective model for prediction based on simulation results. As per the analysis the ANN (Neural Network Cascading) model and DLR could be tested in the real time, integrated model could be developed.