Time Series Feature Extraction

Modified on Fri, 22 Sep 2023 at 01:48 PM


Create new columns of features x’(t) by performing a transformation on an existing column(s) of a time dependent variable x(t), where the time t is also included in the dataset as a separate column.


Raw data may need post processing to be more representative of the physics of the problem trying to be modeled e.g. the force on an object is more proportionately linked to the acceleration on that object than the velocity on that object, so you may wish to take the derivative of the velocity first using Time Series Feature Extraction and then use that derivative as a model input. There are lots of potential features that may be useful to include for your problem.

How to use

This function requires an input Dataset to be selected. This Dataset must have Time as a column. If you have multiple different tests in the same dataset, they must be identified by an ID column.

  • Select the Time column, if not already preselected (if your column is labelled Time it will have been automatically selected).
  • Select one or multiple columns in the Columns field that you would like to extract a feature from. One new column will be added for each column selected in this field.
  • Select the Feature or transformation you would like to apply to each of those columns, the available features are:

Adds a column that is the derivative of the selected column.

Moving Average

Smooths data by performing a centered moving average of user defined window size (more information here).
Low Pass Filter
Uses a butterworth filter to get rid of high frequency signals within the data only allowing through frequencies lower than the cut off freqeuncy (thus smoothing).
Root mean square of the data. (const.)
This and all following features are calculating a single scalar value from a time series and therefore create a new column with a single constant value for each time series. This is denoted by (const.).
Mean value of the data. (const.)
Deviation from the mean. (const.)
SkewnessMeasure of the symmetry of the data compared to a normal distribution. (const.)
Measure of if the data is heavy or light tailed in comparison to normal distribution. (const.)
Peak Amplitude
Maximum amplitude of the signal. (const.)
Amplitude Range
Difference between min and max amplitude of the signal. (const.)
Crest Factor
Indication of how extreme the peaks are in a signal. (const.)
Shape Factor
Numerical Measure of the signal shape. (const.)
  • The new column will be added to the right hand side of the dataset and the feature selected will be added as a suffix to the column heading e.g. Selecting Steering Torque as the column and Derivative as the feature, a new column is added called Steering Torque Derivative.
  • You can also optionally choose to customise the suffix by enabling the checkbox and then entering the desired suffix, Der instead of Derivative for example.
  • Finally, you can choose whether to overwrite the existing dataset or Save the dataset under different name.


Following the steps above, a new column "SteeringAngle Der" will be created. The plot below displays both SteeringAngle and SteeringAngle Der values against time.

Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select atleast one of the reasons

Feedback sent

We appreciate your effort and will try to fix the article