Potential of big data in forecasting travel times

Travel time forecasting is an interesting topic for many intelligent transportation system (ITS) services. Increased availability of data collection sensors increases the availability of the predictor variables but also highlights the high processing issues related to this big data availability. The aim of this paper is to analyse the potential of big data and supervised machine learning techniques in effectively forecasting the travel times. For this purpose fused data from three data sources (Global Positioning System vehicles tracks, road network infrastructure data and meteorological data) and four machine learning techniques (k-nearest neighbours, support vector machines, boosting trees and random forest) were used. To evaluate the forecasting results they were compared in-between different road classes in the context of absolute values, measured in minutes, and the mean squared percentage error. For the road classes with high average speed and long road segments, machine learning techniques forecasted travel times with small relative error, while for the road classes with low average speeds and small segment lengths this was a more demanding task. All three data sources were proven to have high impact on the travel time forecast accuracy and the best results (taking into account all road classes) were achieved for the k-nearest neighbours and random forest techniques.

Jaar van publicatie:2015