Vehicular CO emission prediction using support vector regression model and GIS

Publication Type:
Journal Article
Citation:
Sustainability (Switzerland), 2018, 10 (10)
Issue Date:
2018-09-26
Metrics:
Full metadata record
Files in This Item:
Filename Description Size
sustainability-10-03434_Vehicular CO Emission.pdfPublished Version11.97 MB
Adobe PDF
© 2018 by the authors. Transportation infrastructures play a significant role in the economy as they provide accessibility services to people. Infrastructures such as highways, road networks, and toll plazas are rapidly growing based on changes in transportation modes, which consequently create congestions near toll plaza areas and intersections. These congestions exert negative impacts on human health and the environment because vehicular emissions are considered as the main source of air pollution in urban areas and can cause respiratory and cardiovascular diseases and cancer. In this study, we developed a hybrid model based on the integration of three models, correlation-based feature selection (CFS), support vector regression (SVR), and GIS, to predict vehicular emissions at specific times and locations on roads at microscale levels in an urban areas of Kuala Lumpur, Malaysia. The proposed model comprises three simulation steps: first, the selection of the best predictors based on CFS; second, the prediction of vehicular carbon monoxide (CO) emissions using SVR; and third, the spatial simulation based on maps by using GIS. The proposed model was developed with seven road traffic CO predictors selected via CFS (sum of vehicles, sum of heavy vehicles, heavy vehicle ratio, sum of motorbikes, temperature, wind speed, and elevation). Spatial prediction was conducted based on GIS modelling. The vehicular CO emissions were measured continuously at 15 min intervals (recording 15 min averages) during weekends and weekdays twice per day (daytime, evening-time). The model's results achieved a validation accuracy of 80.6%, correlation coefficient of 0.9734, mean absolute error of 1.3172 ppm and root mean square error of 2.156 ppm. In addition, the most appropriate parameters of the prediction model were selected based on the CFS model. Overall, the proposed model is a promising tool for traffic CO assessment on roads.
Please use this identifier to cite or link to this item: