Abstract
Traffic congestion on ordinary roads including national highways remains an important problem in traffic engineering. Although various kinds of big data have been accumulated through the development of ICT in recent years, their utilization is not necessarily sufficient. In this study, traffic conditions are investigated mainly at signalized intersections, which are the main bottlenecks on ordinary roads, using ETC2.0 probe data, and road structure and land use data are organized using other big data such as digital national land information. Integrating and using these multiple data, the factors of traffic congestion on ordinary roads are quantitatively analyzed. Machine learning model was constructed using XGBoost and interpreted by SHAP. As a result, it was confirmed that one-hour traffic volume and other factors affect the occurrence of congestion and left-turn lane length and other factors may lead to the reduction of congestion. In addition, the effect of extending the right-turn lane on reducing congestion was confirmed by predicting the effect of congestion countermeasures focusing on individual signalized intersections.
