Evaluation of ANN Using Air Quality Tracking in Subtropical Medium-Sized Urban City

The rapid developed communications and artificial intelligence technologies lead people to a higher standard of quality of life. Environmental protection and air quality are more concerned as it is necessary for planning of outdoor activities. To track the air quality, monitoring stations and conventional empirical subjective forecast are used. Because air quality data is seasonal time series, machine learning is a good way to assist the prediction by exploring the seasonality patterns. This study aims to see the implementation of a machine learning model to predict the air quality in a medium-sized urban city. We will see the performance of the multivariate artificial neural networks in predicting the future status of different air pollutant concentrations, such as respirable suspended particulate matter in small and medium-sized developing cities. The neural network was trained on hourly data from 2016 to 2020, with dataset split according to different season groups. Methodology mainly includes model building, training, and testing. Macao was selected for the study. A set of meteorological variables is chosen as multivariate inputs, including air temperature, relative humidity, precipitation, boundary layer height, sea level pressure, and wind. Contributions include seeing the performance of a LSTM model to forecast time series with multivariate inputs.