Intelligent road traffic status detection system through cellular networks handover information: An exploratory study

The methods currently used for primary road traffic data collection have prohibitive costs which compromise coverage of the entire transportation network in a city. Failure to collect information from the road traffic stream leads traffic management authorities to rely on an incomplete picture of the traffic status. This study explores a complementary method to gauge the status of road traffic conditions through the use of cellular networks handover count. To test this method, hourly handover counts were obtained in Lisbon, Portugal, from 39 cellular towers in the vicinity of arterial roads that have 12 traffic counters with an average daily traffic size of 20,500 vehicles. An initial correlation analysis proved the existence of a good relationship between handover and traffic volumes. However, the number of vehicles to handovers ratio at different sites can change up to 10 folds, which has limited the expansion of our model to estimate the absolute traffic volumes based on handover counts. Hence we have classified the hourly traffic counts into three categories: high, medium, and low traffic levels using the 50th and 80th percentiles. Then, half of the data was used to build a multinomial logit (MNL) model and to train an artificial neural network (ANN) in order to relate traffic and handover. The other half of the data was used to validate both models. The MNL and ANN models gave an overall correct classification accuracy of 76.4% and 78.1% respectively. Both models outperformed the accuracy of 70.8% obtained from a City‐wide time‐of‐day traffic profile. The results demonstrate the feasibility of handover based models providing better accuracy in capturing site‐specific traffic profile compared with the typical City‐wide time‐of‐day traffic profile. It can be concluded that this study encourages the exploration of the use of cellphone handover information in estimating the road traffic status.