Accuracy Analysis of Machine Learning Methods for Predicting PM Concentration

In this study, machine learning technique was applied by using PM10, PM2.5 and air quality data acquired from Urban Air Monitoring Network and Meteorological data acquired from Automated Synoptic Observing System (ASOS) and Aerosol Optical Depth (AOD), & ANGS;ngstrom Exponent (AE) data acquired from the ground-based Sun-sky radiometer (AERONET) observation network or Satellite data (MODIS). For the determination of the best machine learning (ML) model, four ML techniques such as Multi Linear Regression (MLR), Support Vector Machine (SVM), Random Forest (RF), Deep Neural Network (DNN) were tested and compared accuracy using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R2, Mean Absolute Percentage Error (MAPE). Since the error range can be presented according to the diversity and variability of input data and ML, it is possible to compare the prediction accuracy of each model or determine the optimal prediction model. We also proved the assumption that more accurate results can be obtained by the optimized ML technique having the lowest error rate. The results showed that optimized ML model has the accuracy of 81.27% for PM10 concentration prediction and 73.25% for PM2.5 concentration prediction. It is expected that expanded air quality information through the using of ML based PM concentration prediction with the remote sensing data.