An Intelligent Framework for Long-term and Short-term Air Pollution Forecasting Using Deep Learning Model

Abstract

Air is an important component of human life, without air life cannot exist. The contaminated air is called pollution of air. Air quality refers to whether the air is bad or good, bad quality air is considered polluted air. Air pollution is caused by a large number of dust particles that are spread in the air and these particles are measured by Particulate Matter (PM), their units in µg/m³ (microgram per meter cube). Traditional statistical forecasting models are computationally expensive and in the case of the large dataset, it takes plenty of time to generate more accurate results. Previous proposed framework was not customizable to an extent that it could easily provide a future long-term prediction of air pollution i.e., PM of 6 hrs., 9 hrs., 18 hrs., 36 hrs. and so on. The deep learning models reduce the computation complexity, consume less time, increase accuracy, and are less error prone as compared to conventional statistical probabilistic models. This proposed framework provides long-term and short-term air pollution forecasting by using a deep learning model. The proposed framework provides accurate forecasting of air pollution in distributed locations. Through validation, it has been observed that the proposed framework provides better results on trendvariant datasets. This framework is reliable enough to provide instantaneous predictions at weather stations about air pollutants based on historical and future data. The proposed framework shows a 3% loss in short-term and 5.5% loss long-term prediction on Punjab EPA Dataset, 2.1% loss in short-term and 3.1% loss long-term on Sofia Air quality dataset (Location 1), 1.2% loss on short-term and 1.1% loss long-term on Sofia Air quality dataset (Location 2) datasets