Prediction of Particulate Pollution (PM10) During the Transboundary Episodic Haze Events in Malaysia Using Machine Learning Approaches Samsuri Abdullah (a*), Aimi Nursyahirah Ahmad (a), Amalina Abu Mansor (b), Mohammad Fakhratul Ridwan Zulkifli (a), Suriani Mat Jusoh (a), Ku Mohd Kalkausar Ku Yusof (b), Marzuki Ismail (b)
a)Faculty of Ocean Engineering Technology and Informatics, Universiti Malaysia Terengganu, 21030, Kuala Nerus, Terengganu, Malaysia
*samsuri[at]umt.edu.my
b)Faculty of Science and Marine Environment, Universiti Malaysia Terengganu, 21030, Kuala Nerus, Terengganu, Malaysia
Abstract
This study investigates the haze episode in Malaysia that occurred from January to February and June to August every year. Data years 2005 until 2014 were utilized from the Malaysia Department of Environment (DOE). The parameters used in this study were the particulate matter with an aerodynamic diameter of less than 10 micrometer, ambient temperature, relative humidity, wind speed, ground-level ozone, nitrogen oxide, nitrogen dioxide, carbon monoxide, and sulphur dioxide to gain a better picture of PM10 variability during transboundary haze. Results showed that the maximum PM10 concentration was higher during transboundary haze events in the years 2013 and 2014. CO was strongly and positively correlated with haze due to its higher emission through large-scale biomass combustion from neighboring countries. ANN model was selected as the best-fitted model in this study. It has lower error measures of RMSE (0.07326) and higher accuracy measures in terms of the correlation coefficient (0.6737) with the optimum number of neurons in the hidden layer is 16, while MLR model has the Root Mean Square Error (RMSE, 126.728), and correlation coefficient (0.445) and PCR model with a value of RMSE (261.471), and correlation coefficient (0.445). The results will help the authorities in getting early information for preserving the air quality, especially during transboundary haze episodes.
Keywords: Haze- Particulate Matter- Artificial Neural Network- Principal Component Regression- Multiple Linear Regression
