Vehicle emission prediction using remote sensing data and machine learning techniques

Abstract

More researchers are using remote sensing technology to measure real-world, on-road automobile emissions of nitric oxide (NO), one of the most important and frequently studied pollutants. Partnered with the National Institute of Water and Atmospheric Research (NIWA) in New Zealand, we aim to establish a robust NO emission factor prediction model using remote sensing data to forecast future emissions. We have conducted this research using real-world data that were collected over a 11-year span between 2005 and 2015. The experimental results have shown that the vehicle emission patterns are continuously changing and the relevance of remote sensing data for future predictions decays as they get older. We propose a three-step machine learning approach to establish this model. We use quantile regression forest (QRF) as the base algorithm and use random forests variable importance measure to validate and interpret the features. We have found empirically, the model is more accurate than models that are based on three other algorithms: linear regression, linear model based recursive partitioning, random forest. Lastly, we have extracted human-interpretable prediction rules from our quantile regression forest based model, using the decision based rule extraction algorithm. The rules are useful to generalise prediction logic from a black-box model such as our QRF based model.