Time Series Forecasting of Air Pollution using Deep Neural Network with Multi-output Learning

Abstract

The main objective of multi-output learning for environmental data engineering is to simultaneously forecast multiple variables for a given input. It has a vital role in ecological decision-making strategies due to the complex features involved. It has an immense impact on model stability for environmental data modeling, especially for air pollution modeling. In recent years, multi-input and multi-output learning has drawn massive attention from researchers and policymakers for air quality modeling and forecasting. Traditional air pollution forecasting models perform multivariate forecasting, which considers multiple variables as input to perform PM2.5 forecasting. Particulate matter PM2.5 and PM10 are both the primary source of air pollution, cause much death worldwide. So, it is crucial to predict both the pollutants simultaneously for a better decision-making process. This research study fills this research gap by developing a multi-output pollution forecasting model to help long-run decision strategies for policymakers and the public. The experiment is conducted for Gucheng location, one of the polluted cities of Beijing, and the experimental results demonstrate its effectiveness in air pollution forecasting.