A Comprehensive Review Of Deep Learning Models For Predicting Air Quality

Abstract

Air pollution affects the sustainability of the ecosystem and health of the population, requiring accurate and reliable air quality forecasting models. Non-linear patterns within air pollution data sets are often challenging for conventional models, including machine learning models, to tackle. Owing to their capability of learning higher-level representations without manual intervention and mimicking long-range dependencies without requiring manual formulation, deep learning models are progressively gaining popularity in air quality forecasting models. Focusing on air pollutants that play significant roles in air quality forecasting models, including PM2.5, PM10, NO₂, SO₂, CO, O₃, this paper presents a comprehensive review of air quality models developed on deep learning models. Detailed descriptions of convolutional neural networks, Additionally included are recurrent neural networks, auto encoders, hybrid models, attention models, long short-term memory networks, and gated recurrent unit networks. Areas such as data sources, preprocessing strategies, and different performance estimation metrics are also presented. Along with this, some of the significant themes, such as missing data, interpretability, and sensor variability, are also addressed. Moreover, some aspects of transfer learning, spatiotemporal learning, and real-time air quality forecasting are mentioned, as well as many themes that must be considered in the context of deep learning algorithms for air quality forecasting. For researchers and individuals who are looking for ways in which accurate air quality prediction models can be created using deep learning approaches, this paper is presented.