Traffic Characteristics of Short and Long Public Holidays: A Hybrid Holiday-Oriented Speed Prediction Approach via Feature Engineering

Atilgan I., Turkmen H. İ., GÜVENSAN M. A.

IEEE Sensors Journal, cilt.23, sa.20, ss.25016-25025, 2023 (SCI-Expanded) identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 23 Sayı: 20
  • Basım Tarihi: 2023
  • Doi Numarası: 10.1109/jsen.2023.3312189
  • Dergi Adı: IEEE Sensors Journal
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Communication Abstracts, Compendex, INSPEC, Metadex, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.25016-25025
  • Anahtar Kelimeler: Feature engineering, historical average (HA), long-short term memory (LSTM), short/long holiday characteristics, support vector regression (SVR), traffic speed prediction
  • Yıldız Teknik Üniversitesi Adresli: Evet

Özet

Special holidays differ from regular week(end) days in terms of traffic characteristics due to spatio-temporal bursts within the city. This results in the time spent in traffic during holidays becoming too unpredictable for many people. In this study, it was revealed that public holidays contribute significantly to the overall traffic speed estimation problem and therefore should be handled separately. Contrary to other studies in the literature, in order to achieve successful results in all long and short holidays, a hybrid holiday-oriented approach that combines the support vector regression (SVR) algorithm and historical average (HA) method is proposed. We additionally feed the proposed model with three novel feature engineering strategies in order to make the system learn similar holidays' characteristics. To ensure the system's effectiveness across a broad geographic scope, training and testing were conducted on 441 road segments located in Istanbul, Turkey. The results demonstrate that the proposed method could achieve up to 29% improvement in terms of mean absolute percentage error (MAPE) values for holiday times over 441 different locations. Moreover, with the help of our novel approach, long-term speed estimation models exceed their limits and we could end up with an average minimization of 2% in terms of mean absolute error (MAE) and MAPE traffic speed prediction error over the year.