Estimation of Air Quality Index by Merging Neural Network with Principal Component Analysis

Abstract

The objective of this study is to merge the Principal Component Analysis (PCA) with Artificial Neural Network (ANN) techniques to get the precise results in estimating the Air Quality Index (AQI). This study is applied for Ain Sokhna city for year 2014. Principal components score (PCs) of 10 historical air quality and meteorological parameters are generated and selected appropriately as the input variables in ANN models for predicting AQI. ANN and Multiple Linear Regression (MLR) models are applied and compared for four methods to estimate the AQI. Ten air pollutant and metrological parameters (actual data) are used as input data in ANN and MLR for creating the first method. The other three methods (Varimax, Equamax and Quartimax) are generated by rotating the actual data using PCA to reduce the ten parameters into four PCs for each PCA method. Consequently, the four PCs of each method are used as input data in ANN (PC-ANN) and MLR (PCR) models. The performance of the methods has been compared using Coefficient of Determination (R 2 ), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). Considering the complexity of the air pollution phenomenon, it is found that the ANN model gives the best solution for all rotating PCs using Varimax method. The results show that the estimation of AQI by merging ANN with PCA (PC-ANN) gives more accurate results than merging MLR with PCA (PCR). This indicates that the ANN method can be applied successfully as a tool for decision making and problem solving for better atmospheric management.