Feature-based time-series classification

Here, we consider a time series as a sequence of its segments approximated by parametric models (e.g. autoregressive model, discrete Fourier transform, discrete wavelet transform). The parameters of the approximating models are used as time-series' features.
Then, we generalize this approach and use the distributions of the parameters estimated for models approximating different time-series' segments.

The proposed approach is applied to the problem of human activity recognition from accelerometer data.

Reference

M. E. Karasikov, V. V. Strijov, Feature-based time-series classification, Inform. Primen., 2016, Volume 10, Issue 4, 121–131.
URL: http://www.mathnet.ru/php/archive.phtml?wshow=paper&jrnid=ia&paperid=452&option_lang=eng
DOI: 10.14357/19922264160413
pdf: full text

Matlab code

Matlab code resides in code.

Python code

All code rewritten in Python for reproducing experiments from the paper can be found in code/python.

Interactive human activity recognition application

An interactive online application is running on http://www.karasikov.com/activity. The source code resides in code/activity_prediction.

Server

The full code of the demonstration server can be found in code/activity_prediction/server.

Android client application

To get the full code of the android application, unpack ActivityPrediction.rar in code/activity_prediction/android.