On Strategies Improving Accuracy of Speed Prediction from Floating Car Data (FCD)

Tüydeş Yaman H., Kocamaz K., Tuncay K., Çınar B.

https://ace2020.org/en/, İstanbul, Turkey, 6 - 08 September 2021, vol.1, pp.1425-1432

  • Publication Type: Conference Paper / Full Text
  • Volume: 1
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.1425-1432
  • Middle East Technical University Affiliated: Yes


For smart mobility, speed data extracted from Floating Car Data (FCD) plays an important role in speed prediction

accuracy. However, there are reliability issues for commercial FCD due to processing of individual vehicle

tracking data, and imposed temporal averaging to compress data size. Furthermore, spatial discretization

significantly affects the accuracy of the prediction due to uneven segment lengths and highly variable data

availability in the network. In this study, these issues are examined in detail, and several strategies to improve

average speed prediction are proposed. An extensive FCD data from a 75-km long corridor is utilized in the

calculations. Firstly, for data reliability, several filters are applied to clean data, then, a robust algorithm is applied

to smoothen the speed data. Secondly, to investigate and reduce prediction errors due to spatial segmentation, a

number of segmentation approaches are developed, and their effects on the average speed prediction are assessed.

Finally, several autoregressive prediction models are implemented and a comprehensive comparison of results is