DEEP LEARNING-BASED AIR POLLUTION PREDICTION IN SMART CITIES

Abstract

One of the most significant issues affecting urban living is air pollution, a form of environmental hazard. One of the air pollutants responsible for a number of illnesses is PM2.5, or Particulates with a diameter of less than 2.5 mm. Therefore, it is crucial to accurately forecast PM2.5 concentrations to protect the public from possible negative consequences of air pollution in advance. It is now feasible to anticipate quality of air in cities using deep learning approach because to the quick advancement of deep learning technology and their use in all facets in life. In the context of smart cities, this study examines experiments including deep learning techniques. Using a deep learning solution, CNN Bi-LSTM, with a spacio temporal feature, and historical features on pollutants, weather, and PM2.5 concentration at nearby stations, it is possible to predict the hourly prediction of air quality in India, especially Delhi. Additionally comparing the performance differences across several algorithms like LSTM, Bi-LSTM, GRU, Bi-GRU, CNN LSTM, and CNN-Bi-LSTM model. The experiment's findings indicate that the mean absolute error of CNN Bi-LSTM and the RMSE both experience drops of 0.20224 and 0.33256, respectively. CNN Bi-LSTM gives the greatest results for air pollution prediction when compared to Bi-LSTM and LSTM, and its training time is cut by 12.5 and 50.9 cent, respectively. . The primary conclusions of this study are as follows, based on performance assessment and result comparison:

this model can successfully capture the spatial and time aspects of the data using CNN and Bi- LSTM, with more precisely and has stability.