Real Time Attention Based Bidirectional Long Short-Term Memory Networks for Air Pollution Forecasting

Abstract

Approximately 95% of the world's population live in places with unsafe air which leads to serious health hazards for people worldwide. Monitoring and preserving air quality has become one of the most essential activities in many industrial and urban areas today. In view of increasingly alarming environ-mental pollution problems due to rapid urbanization, ambient air pollution prediction is becoming extremely important. This paper presents an online real time air pollution prediction system at five prominent locations in Delhi utilizing the past historic air quality and meteorological data. We propose a novel end to end sequential modelling framework to predict the air quality by estimating the concentration levels for various pollutants (nitrogen dioxide (NO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> ), particulate matter (PM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2.5</sub> and PM <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">10</sub> ) and thereby classifying the threat level for them in the next 24 hours. The problem of air pollution prediction system becomes extremely challenging in the presence of unreliable and missing data entries. We also quantify the variations in the pollution concentrations on seasonal basis. We exhaustively evaluate and demonstrate that the proposed approach outperforms several baseline methods for forecasting air pollutants concentration based on data sources obtained in Delhi.