Exploring the reliability of Floating Car Data (FCD) through penetration rate prediction in urban contexts

Monitoring and predicting traffic conditions is a crucial task for transportation agencies. Recent technological advances and the rise of big data have enabled real-time, high-frequency traffic data collection through Floating Car Data (FCD), which offers broader coverage and lower costs compared to traditional methods like fixed sensors. However, FCD is limited as it represents only a sample of users with heterogeneous market shares and, in some cases, lacks vehicle classification information. This study aims to assess the reliability of FCD through a comparative analysis using fixed radar sensors as a ground truth. The analysed variables include vehicle counts to measure FCD Penetration Rates (PRs) as a performance metric and vehicle speeds to assess possible bias phenomena. Additionally, we developed a PR prediction model, identifying the most influential variables through feature engineering and assessing the model's accuracy with Symmetric Mean Absolute Percentage Error (SMAPE). The case study focuses on the city of Catania, Italy, with sensor data obtained from a traffic monitoring system consisting of several counting sections installed along a cordon surrounding the urban area, while FCD were extracted from TomTom portal. Results show spatial and temporal variability in FCD coverage, particularly low PRs at night, and an underestimation of speeds by FCD. The developed predictive model uses widely available FCD data to estimate PRs, helping identify FCD's opportunities and limitations for a more comprehensive understanding of road network performance. Future research will extend the analysis period and integrate more data sources to enhance traffic prediction accuracy and reliability.