Deep Insight into Urban Air Quality Utilizing Neural Networks for Enhanced Prediction in Korean Cities Where Factories and Ecosystem Environments Coexists

Abstract

Increased attention is being given to air pollution in recent times. This study investigated and analyzed particulate matter data from Yeosu, Gwangyang, and Suncheon in Jeollanam-do, with a particular focus on PM2.5. Descriptive statistics, box-and-whisker plots, correlation matrices, time variations, and trend analyses were performed for this purpose. Additionally, a prediction model for PM2.5 concentrations was developed using machine learning techniques, through which future changes in air quality were forecasted.

Calculations were performed using R-based programs and R packages. Hourly PM2.5 data were obtained from air quality monitoring sites in Yeosu, Gwangyang, and Suncheon. After data preprocessing, the optimal prediction model was constructed using Random Forest and Gradient Boosting Machine from various machine learning algorithms.

The research results showed that there was more PM2.5 pollution in Gwangyang compared to Yeosu and Suncheon. The PM2.5 concentrations varied significantly across each monitoring site. Among the monitoring sites, the Yeosu site showed a higher correlation in PM2.5 with each other than other sites. Late winter and early spring showed higher PM2.5 concentrations, while summer and autumn showed lower concentrations. Weekly PM2.5 concentration fluctuations were not significantly different. Daily fluctuations showed an increase in PM2.5 concentrations during times of traffic congestion and a decrease in the afternoon. During the research period, the trend of PM2.5 concentration was generally decreasing.

The accuracy of the prediction model through machine learning was over 90%, and it is expected to assist in establishing effective response strategies for future changes in air quality. This study provided an updated and useful evaluation of recent PM2.5 air quality in Yeosu, Gwangyang, and Suncheon in Korea.