Support vector machine (SVM) and multilayer perceptron (MLP) were used to forecast hourly tropospheric ozone concentration at three locations of Quang Ninh, namely Cao Xanh, Uong Bi and Phuong Nam. Data used to train the models are the hourly concentrations of gaseous pollutants (O3, NO, NO2, CO) and meteorological parameters including wind direction, wind speed, temperature, atmospheric pressure, relative humidity measured in the 2016. Both models accurately forecast tropospheric ozone levels compared to the observation data. The correlation coefficients (r) of the models applied for the three locations range from 0.85 to 0.91. In addition, SVM exhibits a more accurate prediction than MLP, especially for those with large variations, i.e. high standard deviations.

In this paper, three machine learning models have been applied to predict and fill in the missing monitoring data of air quality for Gia Lam and Nha Trang stations in Hanoi and Khanh Hoa respectively, including Autoregressive Moving Average (ARMA), Artificial Neural Network (ANN), and Support Vector Regression (SVR). Two air pollutants being NO2 and PM10 were selected for this study. The experimental results showed that the performance of all three studied models is better than that of some traditional approaches, including Multiple Linear Regression (LR) and Spline interpolation. Besides that, ARMA, ANN and SVR can capture the fluctuation of concentrations of the selected pollutants. These results indicated that the machine learning is a feasible approach to deal with the missing of data which is one of the biggest problems of air quality monitoring stations in Viet Nam.