Deep-Learning-Based Prediction System for Ultrafine Particulate Matter (PM0.1) Concentration Using Meteorological Factors
Abstract
Ultrafine particulate matter (PM0.1) is a global and significant environmental issue because it can deeply translocate the human body, causing adverse health effects and leading to a high mortality rate. This study investigates the relationship between meteorological factors and PM0.1 concentrations, providing insights into the formation and distribution of ultrafine particles. However, accurate measurement of the PM0.1 concentration information is challenging due to sophisticated processes and expensive instruments that make it difficult to access. This study addresses these concerns with a new deep-learning regression model for PM0.1 concentration prediction based on meteorological factors. The model is designed and developed to explore the optimal model structures (hidden layers and neurons) to achieve standard laboratory-based PM0.1 measurement. The model structures are verified by root mean squared error (RMSE) and coefficient of determination (R2) based on predictive performance to prove the laboratory-based standard's accomplishment. The results demonstrate that the proposed model can estimate PM0.1 concentration with high performance, R2 = 92.52%, and RMSE = 0.26 µg/m3, which is precise and reliable for an intelligent-driven PM0.1 concentration prediction system to support preventive health decision-making. This approach contributes to a more comprehensive understanding of atmospheric composition by enabling widespread monitoring of PM0.1, a critical but often unmeasured component of air pollution.