The Effectiveness of Machine Learning Forecasting of PM2.5

Abstract

The purpose of this research was to compare the efficiency of machine learning techniques to predict small dust particles in the air (PM_2.5). The 16 variables were used to construct PM2.5 dust models from air quality and meteorological measurement stations of the Pollution Control Department that was collected an hourly format during June - December 2021, a total of 4,608 rows. The data analysis process is to build a trend prediction model of small dust particles in the air with data estimation techniques (Regression Model). Four different machine learning techniques were used as linear regression, neural networks, support vector machines, and deep learning. The results show that the most suitable technique for modeling dust particles in the air prediction is the neural network technique with the least mean squared error (MSE) of 8.673, the mean squared error (RMSE) of 2.945, the absolute error (AE) of 2.126, the doubled squared error (SE) of 8.709, and the multiple decision coefficient (R²) of 0.948. The results can be used to build a suitable model for trend prediction of small dust particles in the air and set a standard scale for fine particulate matter in the air. Including, the relevant organizations can access the information and used to determine measures to monitor the occurrence of small dust particles. The results can also be further developed an information system for predicting small dust particles to appear in the air more accuracy.