Dynamical Clustering of Time Series Data Using Multi-Decoder RNN Autoencoder

Daisuke Kaji, Kazuho Watanabe, Masahiro Kobayashi

Keywords: Dynamical system, Recurrent neural network, Autoencoder, Variational Bayes, Clustering, Time series data, Driving data

TL;DR: Novel time series data clustring algorithm based on dynamical system features.

Abstract:

Clustering algorithms have wide applications and play an important role in data analysis fields including time series data analysis. The performance of a clustering algorithm depends on the features extracted from the data. However, in time series analysis, there has been a problem that the conventional methods based on the signal shape are unstable for phase shift, amplitude and signal length variations.
In this paper, we propose a new clustering algorithm focused on the dynamical system aspect of the signal using recurrent neural network and variational Bayes method. Our experiments show that our proposed algorithm has a robustness against above variations and boost the classification performance.

Code: https://github.com/nosorog3/MDRA