Novel hybrid model for daily prediction of PM₁₀ using principal component analysis and artificial neural network

Prediction of air pollutants in particular those related to PM₁₀ has developed a huge interest in recent years, mainly due to its impact on environment and humans. There are a large number of factors that influence air pollutant prediction. The researcher has to select the most relevant one by combining different input variables combinations in order to find the combination that provides the best prediction by artificial neural network (ANN). In this work, applications of principal component analysis (PCA) are presented to solve the problem of selection of variables in the prediction of daily PM₁₀. This method is tested by utilizing time series data of solar radiation, vertical wind speed, atmospheric pressure, PM_2.5, benzene, NO and PM₁₀ for Varanasi, India. The results obtained shows that PCA-ANN predicts daily PM₁₀ with mean absolute percentage error (MAPE) of 9.88% and it predicts better than multiple linear regression models.