Prediction Of PM10 Concentration Using Multiple Linear Regression And Bayesian Model Averaging

Ambient PM10 (particulate matter with an aerodynamic diameter less than 10μm) is one of the pollutant that has negative impacts on human health and environment. It is influenced by weather and gaseous parameters. This study is to predict particulate matter (PM10) concentration by using multiple linear regression and Bayesian model averaging. Four stations were selected for three years (2013 until 2015) which are located in Jerantut , Nilai, Seberang Jaya and Shah Alam. Before the analysis, the data was divided into two categories which are training data and validation data. The training data is 70% of observed data (beginning on day 1 until day 255) used to obtain the model. Another 30% of observed data (beginning on day 256 until day 365) were used for validation purpose. The descriptive analysis showed that in 2015, Nilai recorded the highest mean value of PM10 concentration compared to other stations while the highest maximum value of PM10 concentration was recorded at Seberang Jaya station that happened in 2015 due to inter-monsoon season that indicate PM10 level is above threshold value following Malaysia Ambient Air Quality Guideline (MAAQG). To obtain the parameters that contribute to air pollutant for the prediction of particulate matter for the next day (PM10,D1), the training data was analysed using SPSS software for multiple linear regression model and R Software for Bayesian model averaging. The results showed that Shah Alam station is contributing the main parameters which have highest value of adjusted 2 by using multiple linear regression models. Assessment of model performance indicated that Bayesian model averaging (BMA) is the better model to predict PM10 concentration for the next day (PM10,D1) by using the validation data