Analysing several time series at once, from

Tribe3DTribe3D Member Posts: 3 Contributor I
edited January 2019 in Help
I have several time series (not a live stream),  that are pulled in from a single experiment, and that should cross correlate. Specifically they are EEG (brainwave) readings from several sensors. So there are 12 sensors pulling in 6 seconds of data simultaneously. This 6 second section is characteristic of a certain action and gives a specific binary target class, or label. There are 150 training blocks of this data, and 150 to be tested. My question, is how to find a way to input this data into one of the potential ML functions for analysis, since it doesnt appear as a standard table. Additionally, signatures or features, would be expected along the time dimension (like a time series) but also across the 12 sensors, since their grouping also gives a specific signature. Many thanks


  • Telcontar120Telcontar120 Moderator, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,635 Unicorn
    Take a look at the Windowing operators in the new Time Series release (part of the general Modeling operators since RapidMiner 9.0).  You can use them to transform your data to make it suitable for classic machine learning algorithms, although you will need to make some decisions about how large a window to use, how many attributes to include, etc.

    Brian T.
    Lindon Ventures 
    Data Science Consulting from Certified RapidMiner Experts
  • Tribe3DTribe3D Member Posts: 3 Contributor I
    Thanks Brian

    For each block of data to be tested (against a label, or target) there are 12 channels of 240 data points (samples).  So there is a signature (lengthwise) along the wave, as in a non-streamed time series. And they would also correlate accross the samples. So considering the example given in the Windowing operator regarding gas stations, it might be equivalent to 12 gas stations (gas delivered) simultaneously sampled across a 24 hour period , with 240 samples (ie 10 per hour). Many thanks
Sign In or Register to comment.