Intelligent machine learning modelling for air quality index prediction

Abstract

Air quality research affects global warming, climate change, health effects, and urban planning. Predicting air quality status is complex, with increasingly sophisticated monitoring devices for various different gases and other components. Air quality measurements contribute to broader socio-economic development factors, in addition to direct environmental and healthcare impacts. Many methods have been used by researchers to present air-quality levels, reflecting different disciplines and different national standards. This work aims to develop a model to predict the air-quality index, which measures air pollution levels, in order to support healthcare in congested areas. This research presents machine learning models and techniques to predict air quality levels in cities, and to provide accurate measures to support data driven decision making in various sectors aligned with sustainable development, economic growth, and social values. It supports air quality policies formulation with a future vision to eliminate global related consequences, save the world from the pollution and to close the gap in air quality index standardization, with an emphasis on sustainable urban development. This study presents the experimental multivariate Deep Neural Network model and Markov switching model as part of research to develop a hybrid (DNN and Markov) air quality prediction model, with appropriate accuracy attainment, in order to support decisions with timely air-quality measurements by representing the output or air quality levels using neuro-fuzzy logic. DNN-Markov modelling techniques are used to predict air quality, with comparative analysis of locations in Jordan and the UK. Multivariate time series analysis of Big Data from traffic-heavy locations was used, based on environmental conditions at peak hours, giving a highly accurate prediction of the air-quality index for the next hour at a given location, under specific environmental conditions. The air quality index was represented using Neuro Fuzzy Logic as a method to contribute in air quality index predictions within blurry (boundary) values. The selected DNN-Markov hybrid model could predict air quality with accuracy of around (RMSE 7.86) for the location in England, and around (RMSE 15.27) for the one in Jordan.