A hybrid model for forecasting of particulate matter concentrations based on multiscale characterization and machine learning techniques

Accurate prediction of particulate matter (PM) using time series data is a challenging task. The recent advancements in sensor technology, computing devices, nonlinear computational tools, and machine learning (ML) approaches provide new opportunities for robust prediction of PM concentrations. In this study, we develop a hybrid model for forecasting PM10 and PM2.5 based on the multiscale characterization and ML techniques. At first, we use the empirical mode decomposition (EMD) algorithm for multiscale characterization of PM10 and PM2.5 by decomposing the original time series into numerous intrinsic mode functions (IMFs). Different individual ML algorithms such as random forest (RF), support vector regressor (SVR), k-nearest neighbors (kNN), feed forward neural network (FFNN), and AdaBoost are then used to develop EMD-ML models. The air quality time series data from Masfalah air station Makkah, Saudi Arabia are utilized for validating the EMD-ML models, and results are compared with non-hybrid ML models. The PMs (PM10 and PM2.5) concentrations data of Dehli, India are also utilized for validating the EMD-ML models. The performance of each model is evaluated using root mean square error (RMSE) and mean absolute error (MAE). The average bias in the predictive model is estimated using mean bias error (MBE). Obtained results reveal that EMD-FFNN model provides the lowest error rate for both PM10 (RMSE = 12.25 and MAE = 7.43) and PM2.5 (RMSE = 4.81 and MAE = 3.02) using Misfalah, Makkah data whereas EMD-kNN model provides the lowest error rate for PM10 (RMSE = 20.56 and MAE = 12.87) and EMD-AdaBoost provides the lowest error rate for PM2.5 (RMSE = 15.29 and MAE = 9.45) using Dehli, India data. The findings also reveal that EMD-ML models can be effectively used in forecasting PM mass concentrations and to develop rapid air quality warning systems.