Evaluating the Predictive Performance of LSTM and PSO-LSTM Models for Air Quality Forecasting

Abstract

This study compares the performance of LSTM and PSO-LSTM models in predicting air quality based on PM10, SO2, CO, O3, and NO2 parameters. Two main evaluation metrics, MAPE and RMSE, are used to measure prediction accuracy. Prediction accuracy compared to the conventional LSTM, especially on CO and SO2 parameters, where the MAPE and RMSE values were significantly reduced. While the improvements on other parameters such as PM10 and O3 were relatively small, hyperparameter optimization using PSO proved effective in improving the overall prediction performance. Therefore, PSO-LSTM is a superior choice in predicting air quality, especially when faced with more complex data.