Event characteristics that disrupt transport system’s balance

The life of the city is often reflected in traffic patterns: popular sporting events draw crowds, holidays create disruptions, protests may result in road closures, etc. Decades of research on travel demand and network modelling already provide satisfying predictive tools. However, the main research focus has been on regular behaviour, such as peak/off-peak cycles, regular functioning of the infrastructure, and normal weather conditions. Consequently, non-recurrent events severely challenge such models. Under non-recurrent circumstances, the typically expected correlation structures (e.g. between demand flows in neighbor areas; between current and recent values of traffic speeds or travel times) are drastically affected, severely affecting predictions. It is therefore necessary to take into consideration data from different sources. The objective of this research is the development of a methodology that correlates high taxi demand observations with popular events retrieved from Social Media platforms. Using NYC taxi trips public dataset, the average demand of the day was determined using kernel density analysis. Days that showed significant outliers compared to the average day were further studied using a dataset of around 116000 events. The second dataset was retrieved from the Web for the same 6 months period through the direct use of APIs. The correlation step includes the comparison of spatial and temporal kernel density depiction of taxi pick-up locations and events retrieved details. Through the correlation evaluation of traffic data and semantic information, conclusions were made on how the demand of taxi pick-ups changes based on certain event
characteristics.