Benchmark Comparison of Methods for Approximate Neighborhood Estimation of Road Segments in Large-Scale Traffic Networks

Abstract

The accuracy of a traffic prediction model used to predict the future state of a link (i.e. road segment) of interest in a traffic network, can be enhanced by integrating the traffic information of adjacent links. The approximate estimation of such a neighborhood of links for all links of the traffic network is a type of approximate k-nearest neighbor search problem (k-NNS), which can prove to be quite computationally expensive especially for large-scale networks. In this paper, methods for solving this problem in large-scale traffic networks are presented and compared with each other in terms of both accuracy and running-time performance. Experiments have been conducted using real-world map data from the city of Thessaloniki, and the preliminary results indicate the effectiveness of the methods in solving the neighborhood estimation problem.