Hands-On Machine Learning with TensorFlow.js

Constituent frequencies as a feature

In general, the original signal isn't suitable for machine learning algorithms because it is difficult to extract meaningful features from sequential data points. The component of each frequency can represent the original signal. Once the original sequence has been converted into the constituent frequencies, it is handled as a normal vector of features.

There are mainly two advantages to using components as features:

• We can use more tangible elements by converting the time domain into a feature space with a fixed size. This format allows us to provide traditional machine learning algorithms such as decision trees.
• Fourier transformation is a reversible operation. It is possible to construct the original ...