Skip to content

Extract Features

This class analyzes movement data (like accelerometer readings from a smartwatch) by dividing it into time chunks ("windows") and calculating features (mathematical summaries) for each chunk. These features help in understanding patterns in movement, such as average speed, variability, or frequency of movement. The results are saved in a file for further analysis or use in machine learning.

Step-by-Step Explanation:

  1. Input Data:
    • The input is a table containing accelerometer data, which measures movement along three axes: x, y, and z.
    • The data is recorded continuously, so this class divides it into manageable time chunks, called "windows."
  2. Window Creation:
    • Each window represents a specific time span of movement (e.g., 1 minute).
    • A "step size" determines how much the window moves forward for the next chunk (e.g., every 30 seconds).
    • This ensures all the data is analyzed efficiently.
  3. Feature Extraction:
    • The class calculates a variety of features for each window. These features summarize the movement data and make it easier to analyze or use in machine learning.
    • Features are grouped into domains, including:
      • Time Domain Features: Basic statistics like average movement, variability, and extremes.
      • Spatial Features: Measures like angles (tilt) and correlations between axes.
      • Frequency Domain Features: Patterns in how often movement occurs (e.g., fast vibrations vs. slow sways).
      • Statistical Features: Percentiles to capture movement range.
      • Wavelet Features: A way to break movement into simple patterns for analysis.
    • Users can choose which domains to include or calculate all features if no preference is set.
  4. Magnitude Calculation:
    • The class can also calculate features for the "magnitude," which combines x, y, and z movements into one overall measure of intensity.
  5. Combining Features with Labels:
    • If emotion labels (like arousal and valence) are available, they are added to the features for each window. This makes it possible to link movement patterns with emotions.
  6. Saving the Results:
    • The calculated features for all windows are saved in a file. The file name includes details like the window size, step size, and selected domains for easy identification.
  7. Friendly Messages:
    • Messages like "All features extracted successfully" are displayed to let the user know the process is complete.