Data-driven Methods for Identifying Travel Conditions Based on Traffic and Weather Characteristics

Accurate and reliable traffic state estimation is essential for the identification of congested areas and bottleneck locations. It enables the quantification of congestion characteristics, such as intensity, duration, reliability, and spreading which are indispensable for the deployment of appropriate traffic management plans that can efficiently ameliorate congestion problems. Similarly, it is important to categorize known congestion patterns throughout a long period of time, so that corresponding traffic simulation models can be built for the investigation of the performance of different traffic management plans. This study conducts cluster analysis to identify days with similar travel conditions and congestion patterns. To this end, travel, traffic and weather data from the Smart Mobility Living Lab of Thessaloniki, Greece is used. Representative days per cluster are determined to facilitate the development of traffic simulation models that typify average traffic conditions within each cluster. Moreover, spatio-temporal matrices are developed to illustrate time-varying traffic conditions along different routes for the representative days. Results indicate that the proposed clustering technique can produce valid classification of days in groups with common characteristics, and that spatio-temporal matrices enable the development of traffic management plans which encompass routing information for competing routes in the city of Thessaloniki