Deep Unsupervised Multivariate Time Series Classification

Maxim Lunitsin, Vadim Selyutin, Semyon Glushkov, Vladislav Zhuzhel, Liliya Mironova, Kirill Pavlenko

Abstract

Time-series techniques are valuable tools to solve many real-world problems, with classification being a crucial one. In the literature, there are well-established solutions for the supervised setting. However, few entries deal with classification in an unsupervised manner. Similarly, the field of unsupervised classification of time series using deep learning methods remains mostly unexplored. In this work, we propose a method to build a deep learning model for unsupervised time series classification and compare its performance with existing approaches from the supervised and unsupervised literature.

References

1. Asano, Y. M., Rupprecht, C., and Vedaldi, A. Self-labelingvia simultaneous clustering and representation learning. Conference paper at ICLR 2020, 2020. Chen, Q., Hu, G., Gu, F., and Xiang, P.
2. Learning optimal warping window size of dtw for time series classification. 2019 IEEE 15th International Conference on AutomationScience and Engineering (CASE), 2019. Elliot, A. and Hsu, C. H.
3. Time series prediction: Predictingstock price. arXiv preprint arXiv:1710.05751v2, 2017. Graves, A., rahman Mohamed, A., and Hinton, G.
4. Speechrecognition with deep recurrent neural networks. arXiv preprint arXiv:1303.5778v1, 2013. Guijo-Rubio, D., Dur ́an-Rosal, A. M., Guti ́errez, P. A., Tron-coso, A., and Herv ́as-Mart ́ınez, C.
5. Time series cluster-ing based on the characterisation of segment typologies. arXiv preprint arXiv:1810.11624v1, 2018.Wang, J., Chena, Y., Haoc, S., Penga, X., and Hua, L.
6. Deeplearning for sensor-based activity recognition: A survey. Pattern Recognition Letters, 2017.Wang, Z. and Oates, T.
7. Time series classification fromscratch with deep neural networks: A strong baseline. arXiv preprint arXiv:1611.06455v4, 2016.Yang, Q. and Wu, X. 10 challenging problems in datamining research. International Journal of InformationTechnology Decision Making, 5(3):597–604, 2016.