Prediction of the concentrations of PM1, PM2.5, PM4, and PM10 by using the hybrid dragonfly-SVM algorithm

This paper aims to model the daily evolution for particulate matter concentrations of less than 1 μm (PM1), 2.5 μm (PM2.5), 4 μm (PM4), 10 μm (PM10), and PM-Total, based on weather factors (WF), by using the hybrid dragonfly-SVMr algorithm. Hourly data on atmospheric concentrations of PMi and WF were recorded simultaneously at an automatic air quality check station located at an urban site in Algiers, using the fine dust measurement device, Fidas® 200. The number of data collected on PM was 540 measurements. In this study, the meta-heuristic dragonfly algorithm (DA) was used in order to select the optimal hyper-parameters of the Support Vector Machine model. For this, a MATLAB® program based on the dragonfly optimization algorithm coupled with the SVM regression algorithm has been written in order to correlate for the PMi concentrations. The obtained results show that the established model has good predictive performance, with a coefficient of determination R2 = 0.98 and root of the mean square error RMSE = 1.9261.