A Bilevel Traffic Data Extraction Procedure via Cellular Phone Network for Intercity Travel

Abstract

The absence of traffic surveillance infrastructure, in many developing countries, hinders any efforts for dealing with the daily witnessed traffic chaos. The use of the cellular phone (CP) network data for traffic data collection is a promising option for large-scale coverage, given a high CP penetration rate. This article presents a bilevel procedure for the extraction of classified vehicular traffic counts for different vehicle types, in a given roadway segment, for intercity travel. The bilevel procedure operates in an offline setting, independent from any secondary traffic surveillance system. At the first level, cellular phones on board the same vehicle are clustered using a “data swarm clustering” algorithm. At the second level, a genetic fuzzy classifier (GFC) is used for vehicles classification. The development and testing of the proposed procedure was conducted using a traffic/CP simulation platform. At the development phase, the swarm-based clustering algorithm achieved 93% clustering accuracy (vehicle count accuracy). At the second level, the fuzzy classifier successfully classified around 85% of the vehicles. The procedure was further evaluated using a microsimulation model of a major travel corridor in the Greater Cairo Region, Egypt. Superior performance was achieved, at the clustering level, with an accuracy of 97.6%. The revealed accuracy demonstrates the efficiency of the developed procedure for extracting unclassified vehicular counts from CP data. At the classification level, accuracy was reduced to 70.6 ± 11.5%. Achieved classification results are promising from a conceptual perspective. Nevertheless, further investigation is crucial for enhanced classification performance and robustness.