Applying Decision Tree Algorithm for Air Quality Prediction in Bangladesh

Rapid urbanization, socio-economic development, and industrialization result in serious deterioration in air quality (AQ). For environmental management, prediction can be a mitigation and regulation for an adaptation method. This paper applies a supervised machine-learning (ML) technique, decision tree (DT), to predict AQ. In Matlab 2018b, AQ values and climate variables (temperature, relative humidity, wind speed, and rainfall) classify categorical AQ. For eight divisions in Bangladesh, AQ datasets are obtained from the Department of Environment (DOE), while weather variables are acquired from the National Aeronautics Space Administration (NASA)-Prediction of Worldwide Energy Resources (POWER) project. The experiments include daily observations for seven years (2014 to 2020) indicating an average unhealthy AQ (65 to 75% per year) among the chosen metropolitans. DT as a predictive model, datasets from Dhaka are utilized in training (80%) and validation (20%) resulting in an accuracy of 98.8%. This model further is applied to forecast monthly AQ for Chittagong and found predictability ≥97%. Finally, AQ is predicted and found 96% accuracy for eight cities (year: 2020). The investigations encourage providing AQ alerts to the public mostly among data-sparse regions.