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.
