Comparative analysis of Air Quality Index prediction using deep learning algorithms

This paper comprehensively reviews and compares methodologies used to monitor air quality and their impact on human health. With urbanization and industrialization increasing in emerging nations, air pollution levels have become a significant threat to human well-being. The study highlights the importance of reducing exposure to air pollution for the improvement of public health. The paper focuses on the comparative analysis of measuring the Air Quality Index (AQI) using deep learning algorithms like Long Short-Term Memory (LSTM) and classical machine learning models such as Autoregressive Integrated Moving Average (ARIMA), Decision Tree, K-Nearest Neighbour, Extreme Gradient Boosting, Gradient Boosting, Adaptive Boosting, Huber Regressor, and Dummy Regressor for AQI prediction. The performance of these models is evaluated using daily and hourly time series data from 2014 to 2018, with the Root Mean Squared Error (RMSE) used as the performance indicator. The results demonstrate that LSTM outperforms ARIMA, particularly with hourly data. For daily data, ARIMA achieved an RMSE of 97.88, whereas LSTM obtained an RMSE of 143.07. On the other hand, for hourly data, ARIMA yielded an RMSE of 69.65, while LSTM achieved a lower RMSE of 44.6539. These findings highlight the potential of deep learning algorithms, specifically LSTM, in accurately forecasting air quality.