Determination of significant variables to particulate matter (PM₁₀) variations in northern region, Malaysia during haze episodes (2006-2015)

Abstract

The most substantial air pollutant variables during haze episode in Northern region for 10-consecutive years (2006-2015) were analyzed and highlighted. ANN together with SAPCR were integrated to identify the variables contributed to fluctuation of particulate matter (PM₁₀) during haze period. 13 variables including air pollutant and meteorological factor were included as explorable variables. The humidity, wind speed and ozone were recognized as determinant to PM₁₀ variation during haze from 2006-2015. Three artificial neural models were created based on all parameters, leave-out method and PCR-factor loading. The best model will be selected based on a few criterions like determination of coefficient, R², root-mean-square-error and squared sum of all errors. ANN-HM-LO was a better model than ANN-HM-PCR in overall prediction performance with R² result for ANN-HM-LO was 0.839, whilst ANN-HM-PCR was just 0.801.

Keywords: haze studies; sensitivity analysis; artificial neural network; principal component regression