Prediction of the Level of Air Pollution During Wildfires Using Machine Learning Classification Methods
The recent increase of forest fires due to agricultural field burning in the South East Asian region has led to haze episodes in Malaysia which contributed to the increasing number of hospital visits for treatments related to respiratory diseases. With the increase of air pollution, it becomes a necessity to attempt at investigating and predicting the air pollution levels, which would in turn which would lead to proper strategies so untimely effects to human health can be kept at a minimum. The Air Pollutant Index is used to identify and classify the ambient air quality status, However the lack of ground air quality monitors which compute the API generally leads to unreliable warning information. Recent studies indicate that data retrieved from remote sensing satellites is now an emerging alternative for air quality prediction at the ground level, hence this research aims to use satellite-based data to predict the air quality of East Malaysian cities with the help of different classification algorithms. Aerosol Optical data, Meteorological data and Fire data were collected from different satellite sources, two algorithms were selected and modelled. The two algorithms which were implemented, were Random Forest and Gradient Boosting, when trained and validated they both algorithms performed reasonably well with an accuracy 0.89 and 0.85 respectively, for the city of Kuching.