Multi-Scale Temporal Convolutional Networks for Long-Range PM2.5 prediction in Taiwan's Monsoon-Influenced Climate

Abstract

One of the most important components of the environment is air. The earth's temperature, ecosystems, human health, and environmental sustainability are all at risk due to the growing global air pollution catastrophe. Because of how sneaky it is, air pollution has been called a silent killer. Its harmful effects are further highlighted by its indirect impacts on human health. Millions of lives could be saved worldwide if air quality is detected early. Consequently, there is a lot of interest among researchers in the analysis and prediction of air pollution. Neural networks, deep learning, and conventional machine learning are among the research topics. The problem of correctly and efficiently forecasting air pollution becomes crucial. Using ML, DL, and neural network techniques, research aims to forecast and make prediction on PM2.5 concentration at the MCMUG station in Taiwan. Based on the LSTM deep learning method, Random Forest ML algorithm, ANN algorithm, and Gradient Boosted Model which incorporates the XGBoost, LightGBM, and CATBoost techniques? As inputs to the model, the Taiwan Air Quality Monitoring Board's traffic statistics, pollutant information, and climatic characteristics from January 1, 2018, to December 31, 2023 were examined. These are the models beat the predictive performance of existing conventional models. Statistical indicators like RMSE, MAE, and MSE respectively. R2 the co-efficient determinant will be used to evaluate the MCMUG station the effectiveness of various strategies