Prediction of Time Series Data Based on Transformer with Soft Dynamic Time Wrapping

Kuo-Hao Ho, Pei-Shu Huang, I-Chen Wu, Feng-Jian Wang

Abstract

It is a challenge to predict the long-term future data from time series data. This paper proposes to use a Transformer with soft dynamic time wrapping for early stopping criteria, called a soft-DTW Transformer. Our experiment in an open-source dataset HouseTwenty shows that the average prediction error rate with soft-DTW Transformer is 27.79%, greatly reduced from 45.70% for using SVR, a common time series method.
BibTeX
@inproceedings{ho2020prediction,
  author = {Kuo-Hao Ho and Pei-Shu Huang and I-Chen Wu and Feng-Jian Wang},
  title = {{Prediction of Time Series Data Based on Transformer with Soft Dynamic Time Wrapping}},
  booktitle = {ICCE-TW},
  year = {2020},
  doi = {10.1109/icce-taiwan49838.2020.9258155},
}

Loading related content...

Figure