Detail publikace

Towards Robust and Accurate Traffic Prediction Using Parallel Multiobjective Genetic Algorithms and Support Vector Regression

PETRLÍK, J. SEKANINA, L.

Originální název

Towards Robust and Accurate Traffic Prediction Using Parallel Multiobjective Genetic Algorithms and Support Vector Regression

Anglický název

Towards Robust and Accurate Traffic Prediction Using Parallel Multiobjective Genetic Algorithms and Support Vector Regression

Jazyk

en

Originální abstrakt

The support vector regression (SVR) is a very successful method in solving many difficult tasks in the area of traffic prediction. However, the performance of SVR is very sensitive to the parameters setting and the selection of input variables such as sensors providing the input data. In this paper, we describe a new method, which simultaneously optimizes the meta-parameters of SVR model and the subset of its input variables. The method is based on a multiobjective genetic algorithm. The proposed implementation is intended for a parallel environment supporting OpenMP. We evaluated the method in the tasks of data imputation, short term prediction of traffic variables and travel times prediction using real world open data. It was confirmed that the simultaneous optimization of SVR parameters and input variables provides better quality of prediction than previous methods.

Anglický abstrakt

The support vector regression (SVR) is a very successful method in solving many difficult tasks in the area of traffic prediction. However, the performance of SVR is very sensitive to the parameters setting and the selection of input variables such as sensors providing the input data. In this paper, we describe a new method, which simultaneously optimizes the meta-parameters of SVR model and the subset of its input variables. The method is based on a multiobjective genetic algorithm. The proposed implementation is intended for a parallel environment supporting OpenMP. We evaluated the method in the tasks of data imputation, short term prediction of traffic variables and travel times prediction using real world open data. It was confirmed that the simultaneous optimization of SVR parameters and input variables provides better quality of prediction than previous methods.

Dokumenty

BibTex


@inproceedings{BUT119857,
  author="Jiří {Petrlík} and Lukáš {Sekanina}",
  title="Towards Robust and Accurate Traffic Prediction Using Parallel Multiobjective Genetic Algorithms and Support Vector Regression",
  annote="The support vector regression (SVR) is a very successful method in solving many
difficult tasks in the area of traffic prediction. However, the performance of
SVR is very sensitive to the parameters setting and the selection of input
variables such as sensors providing the input data. In this paper, we describe
a new method, which simultaneously optimizes the meta-parameters of SVR model and
the subset of its input variables. The method is based on a multiobjective
genetic algorithm. The proposed implementation is intended for a parallel
environment supporting OpenMP. We evaluated the method in the tasks of data
imputation, short term prediction of traffic variables and travel times
prediction using real world open data. It was confirmed that the simultaneous
optimization of SVR parameters and input variables provides better quality of
prediction than previous methods.",
  address="IEEE Computer Society",
  booktitle="2015 IEEE 18th International Conference on Intelligent Transportation Systems",
  chapter="119857",
  doi="10.1109/ITSC.2015.360",
  edition="NEUVEDEN",
  howpublished="electronic, physical medium",
  institution="IEEE Computer Society",
  year="2015",
  month="september",
  pages="2231--2236",
  publisher="IEEE Computer Society",
  type="conference paper"
}