Application of deep learning techniques for analysis and prediction of particulate matter at Kota city, India

Abstract

Air pollution significantly threatens human health and the environment, making accurate prediction of pollutant concentrations crucial for effective mitigation. This study leverages deep learning models, specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, to predict concentrations of PM₁₀ and PM_2.5. The analysis utilizes hourly air quality data from July 1, 2017, to December 30, 2022, collected from the portals of the Central Pollution Control Board (CPCB) and Rajasthan State Pollution Control Board (RSPCB) for Kota city Rajasthan. Data preprocessing involves cleaning, normalization using a min-max scaler, and handling missing values with Multiple Imputation in XLSTAT. The methodology encompasses dataset loading, preprocessing, and data splitting, followed by model training and evaluation. Python libraries such as Pandas, Numpy, TensorFlow, and Matplotlib are employed for data analysis and visualization. Performance metrics, including Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and R² score, are calculated to assess the models' predictive accuracy. The results demonstrate that GRU model effectively capture temporal dependencies in air quality data, offering reliable predictions for PM₁₀ and PM_2.5 concentrations with  41.85 and 17.73 RMSE values for PM₁₀ and PM_2.5 . These findings underscore the potential of deep learning models in air pollution forecasting, providing valuable insights for policymakers to implement timely interventions.