Investigation of Machine Learning and Deep Learning Approaches for Early PM2.5 Forecasting: A Case Study in Vietnam

Air pollution, a pressing issue in Vietnam, is primarily caused by elevated levels of particulate matter 2.5 (PM2.5). Accurate PM2.5 forecasting is crucial for effective air quality management. This study uses advanced approaches, such as Long Short Term Memory (LSTM-based) models, to estimate PM2.5 levels in Vietnamese locations. Ten locations were initially chosen for prediction, but the pre-processing reduced the number to six due to missing critical values. The study constructs and rigorously tests prediction models using historical meteorological data with conventional statistical indicators. It also explores model combinations for increased reliability and acknowledges persistent problems in improving accuracy and real-world application. In conclusion, this work provides unique insights into PM2.5 forecasting, particularly the superior performance of the LSTM-based model, to aid in mitigating the causes of air pollution.