Chapter 8: Audio Signal Classification with an FFT and a Gradient-Boosted Forest
In this chapter, we will break from forecasting to perform a different type of machine learning on time series data: classification. Using the Fourier transform, we will transform our data and perform dimensionality reduction, then train a familiar classification model with input and target columns to classify an audio source.
The Fourier transform, however, has a myriad of applications in time series analysis beyond classification. It is used to better explore time series data in search of patterns by shifting to the frequency domain where we can view component seasonal patterns. Furthermore, it is used to construct complex state space models capable of incorporating more seasonalities than the SARIMA.
Before we can reach that conclusion though, we will discuss why working with high-frequency time series data can be tricky, introduce the theory behind the Fourier transform, and discuss how window...
