Traffic-Related Air Pollutant (TRAP) Prediction using Big Data and Machine Learning

Abstract

The negative impact of the Increasing air pollution on the global economy, quality of life of humans and health of animals and plants has been enormous. Several works of literature, reports and news around the world have highlighted the risk posed by the ever-in creasing air pollution and the threat to the lives of vulnerable groups such as children, the elderly, and people with respiratory and cardiovascular problems. The closest to home among all the air pollutants are the Traffic-Related Air Pollutants (TRAP), and they contribute the most to the risk posed to global health. This emphasises the urgency of the need for a highly accurate air pollution prediction model. Researchers have been able to achieve significant performance gain in predicting many of the pollutants except for the TRAP such as CO and NO which reported the worse prediction performance in many studies. CO and NO have been among the major pollutants of concern globally as they are linked to critical health hazards. Based on the established urgency of improving the accuracy of pollution prediction models, we collect recent data for six months and at high granularity in terms of time and location. The data is pre-processed and used to develop a Machine Learning (ML)based air pollution prediction model with high granularity and accuracy while focusing on traffic-related air pollutants CO and NO. Using the benchmarks r2and RMSE score, our ML models outperformed that of the studies reported in the literature for the prediction of TRAPs. This in part is due to the high data granularity we considered in terms of time and location.