This question is somewhat inspired by the answer to Features for time series classification.
The difference to that question is that I have a dataset with multi-dimensional time-series where I have several binary-valued features, not continuously-valued features.
Moreover, I do not need to classify the whole series but rather classify the small windows (length 10-100 while the whole time-series is rather of length ~20000).
The question: which of the features mentioned in the attached answer would still apply in the case of binary features?
- Does it make sense to perform some frequency-domain analysis (eg. to know how fast the binary values are changing) and how to choose a window for DFT?
- What about other mentioned features such as skewness, kurtosis and parts of ARIMA model?
