Short-Term Prediction Models Of Pm10 Concentrations In Peninsular Malaysia Using Multivariate Time Series And Machine Learning Methods

• Main Author: Ramli, Norazrin
• Format: Thesis
• Language: English
• Published: 2021
• Subjects: L Education (General)
• Online Access: http://eprints.usm.my/52217/1/NORAZRIN%20BINTI%20RAMLI.pdf; http://eprints.usm.my/52217/

The particulate matter with an aerodynamic diameter less than 10 μm (PM10) is identified as one of the dangerous air pollutants to human health and the concentrations of PM10 in Asian and Pacific cities remain as the most problematic local air pollution issues. The objectives of the research are to determine the characteristics and trend of PM10 concentrations in Malaysia from 1999 to 2015, to propose a Multivariate Time Series (MTS) analysis using Vector Autoregressive (VAR) to predict the short-term PM10 concentrations and interpret the relationship between PM10 concentrations and meteorological parameters using the graphical view of causality. Three models for short-term prediction of PM10 using Multiple Linear Regression (MLR), Bayesian Model Averaging (BMA) and Boosted Regression Tree (BRT) model. The performance indicators (R2, IA, MAE, RMSE, and MAPE) are applied to obtain the best model. A study using seventeen years of air quality monitoring data from the Department of Environment Malaysia (DOE) was used with eight parameters (PM10, NO2, SO2, CO, O3, wind speed, temperature, and relative humidity) and nine monitoring stations were selected which included Kangar, Perai, Shah Alam, Nilai, Larkin, Pasir Gudang, Kertih, Kota Bharu and Jerantut to represent the Northern, Central, Southern and East of Peninsular Malaysia. The trend analysis used the Mann-Kendall test for trend detection and Sen’s slope estimator for trend estimation using monthly average and maximum monthly of PM10 concentrations.