Evaluating Machine Learning Models for PM2.5 Prediction: A Case Study on Air Pollution in Beijing

Abstract

Large pollution impacts on human, animal, and plant health, along with advanced computing technologies capable of managing big data, create new opportunities for applying ML to improve air quality observation. Questions also continue to increase as more are created about how the performance of newer, hybrid ML models is matched to a particular application for the most suitable ML model. This paper presents a systematic review of state-of-the-art studies that implement ML techniques in the context of PM2.5 concentration prediction, focusing on analyzing dataset size, hyperparameters, and preprocessing techniques to answer these questions. This review investigates some proposed ML techniques and models applied in Beijing by highlighting their main characteristics and relevant results. They then pointed out that hybrid models are capable of uncovering the hidden features of data, which was not possible by single approaches with high dimensions. Another conclusion was drawn that air pollution prediction models have to be compared under the same conditions with the same future characteristics.