Improve time series forecast performance with a technique from signal processing Modeling time series data and forecasting it are complex topics. There are many techniques that could be used to improve model performance for a forecasting job. We will discuss here a technique that may improve the way forecasting models absorb and utilize time features, and generalize from them. The main focus will be the creation of the seasonal features that feed the time series forecasting model in training — there are easy gains to be made here if you include the Fourier transform in the feature creation process. This article assumes you are familiar with the basic aspects of time series forecasting — we will not discuss this topic in general, but only a refinement of one aspect of it. For an introduction to time series forecasting, see the Kaggle course on this topic — the technique discussed here builds on top of their lesson on seasonality. Consider the Quarterly