Layman's explanation of when to use which smoother algorithm/technique: FFT, loess, Savitzky-Golay, etc



As an analytics practitioner, I frequently come across noisy data (e.g. IoT data). When building a model or machine learning algorithm, it can be advantageous to smooth this data. Over the years, I have come across a variety of smoothing techniques, but have never been able to synthesize succinctly when to use one technique over another.

I am hoping someone can shed some light on any/all of the following techniques. Even better if the response can be kept at a high-level / layman's explanation to facilitate communication to a non-technical audience.

  • Fast Fourier Transform (FFT)
  • Savitzky-Golay
  • Gaussian Filter
  • Moving average
  • Splines
  • Kernel Smoothing

In particular, are there conditions/situations in which one should not use one or more of the above methods? Feel free to add other methods/techniques to the arsenal.


Posted 2018-11-27T15:42:22.273

Reputation: 130

1Fast Fourier Transform (FFT) is not a smoothing technique. – Brian Spiering – 2018-12-19T16:06:25.920

@BrianSpiering I agree with you, but in this context, I think it makes sense to list it here. Rather than use the full signal, one could use low-pass filters -- here's an interesting application of this technique/approach for doing text sentiment analysis.

– JasonAizkalns – 2018-12-19T16:51:37.827

No answers