Forecasting of CO and O₃concentrations using optimized deep learning: A case study for Kuala Lumpur, Malaysia

Air pollution is caused by the excessive emission of sulphur dioxide, particulate matter, nitrogen dioxide, ozone and carbon monoxide concentration, causing significant threats to environmental health and sustainability. Accurate air quality forecasting is crucial in addressing these issues and promoting a healthier environment. Therefore, this study proposes implementing deep learning architecture and a metaheuristic algorithm for precise air pollutant concentration forecasting. This study uses long short-term memory (LSTM) to learn complex nonlinear time series data and subsequently forecasts air quality for two air quality monitoring sites in Kuala Lumpur, namely Batu Muda and Cheras. The hyperparameters of the LSTM model, such as the number of hidden units in two LSTM layers, batch size and learning rate, are optimized using optimization techniques, namely marine predators algorithm (MPA), to enhance LSTM performance in learning complex temporal patterns and capture long-term dependencies of air quality data for accurate forecasting. Hybrid MPA-LSTM forecasts carbon monoxide and ozone concentration at two target sites. Subsequently, the performance of the model is evaluated in terms of statistical evaluations namely root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R2). The results demonstrate that MPA-LSTM outperforms all baseline models such as multilayer perceptron (MLP), gated recurrent unit (GRU) and LSTM for forecasting carbon monoxide and ozone concentrations at both target locations, indicating the superiority of the model in handling different dataset characteristics.