Creation of a Dust Warning System and a Model for Predicting the Amount of Small Dust Particles PM 2.5 Using Artificial Neural Networks for Local Development

Abstract

This research developed a monitoring system for fine particulate matter (PM2.5) capable of issuing alerts when levels exceed established standards via the Line application and forecasting future particulate levels using artificial neural networks. The forecasting model was created using Rapid Miner Studio and implemented at three locations: Bansomdejchaopraya Rajabhat University in Bangkok, Bang Thorat sub-district in Samut Sakhon Province, and Mae Rim district in Chiang Mai Province. Data were collected from March 1, 2025, to May 31, 2025, for model development and from June 1, 2025, to June 30, 2025, for forecast comparison. The variables included PM2.5 levels, temperature, relative humidity, and wind speed. The analysis revealed that the most accurate forecasts were in Chiang Mai, with the lowest volatility at 0.03% (Mean Absolute Error = 0.155, Mean Absolute Percentage Error = 0.4%). In contrast, Bangkok and Samut Sakhon showed volatilities of 2.31% (MAE = 0.43, MAPE = 5.2%) and 5.54% (MAE = 2.08, MAPE = 6.5%), respectively, highlighting the effectiveness of neural networks in air quality forecasting. The findings from this research can inform the development of local environmental management policies and plans, ultimately improving public health and quality of life.