Air Pollution Prediction Using Long Short-Term Memory Variants

Air pollution is an unresolved environmental problem that poses a health threat in many countries. Nearly 90% of the global population is exposed to polluted air that exceeds established air quality guidelines. Jakarta is one of the most polluted cities in Asia and even the world. The Air Quality Index (AQI) is a measurement index that shows the air quality in an area. The worse the air quality, the higher the AQI value. Jakarta has an average AQI value of 110, which is categorized as unhealthy for some groups of people. For this reason, it is necessary to predict air pollution to take the proper steps to protect their health. One of the deep learning approaches that can handle time series cases is the long short-term memory (LSTM). This research aims to accurately predict air pollution based on AQI using LSTM algorithm variants: Vanilla LSTM, Bidirectional LSTM, and Stacked LSTM. The trained model was evaluated using the root mean square error (RMSE) and mean absolute error (MAE) metrics. Based on the experiment results, the Bi-LSTM model with RMSprop optimizer and 0.0001 learning rate could provide the best results with an RMSE value of 16.68 and an MAE of 12.76. As the best model, Bi-LSTM was implemented to predict Jakarta’s AQI in the following week. The results show that AQI in Jakarta would insignificantly increase.