Travel speed estimation from cellular networks using modified Data Swarm Clustering algorithm

Traffic management operations contribute to alleviate traffic congestion. The absence of traffic surveillance systems hinders the adoption of many traffic management applications. Cellular networks provide cost effective alternatives for traditional traffic surveillance methods. However, a secondary system, which acts as a reference, is needed to extract traffic data. Clustering algorithms as Data Swarm Clustering (DSC) was used to convert cellular based data into vehicular based data without the use of a secondary system. This paper examines the validity of traffic data estimated from vehicular based data extracted from cellular networks merely without the use of a secondary reference system. The procedure is examined on a realistic network in Egypt, 26^th of July corridor, modeled in PARAMICS micro-simulation tool. Harmonic mean speed extracted from cellular networks is compared to that estimated from the vehicles released by PARAMICS, which are considered the ground truth values. The results depicted the out performance of the proposed model especially at low penetration rates. The MAPE was 1.39±1.28% at 100% CPs penetration rate and reached 11.83% at 5% CPs penetration rates.