Feature Transforms
Many models or training processes depend on the assumption that the data is distributed according to the normal distribution. Even the most widely used descriptors, the arithmetic mean and standard deviation, are largely useless if your dataset has a skew or several peaks (multi-modal). Unfortunately, observed data often doesn't fall within the normal distribution, so that traditional algorithms can yield invalid results.
When data is non-normal, transformations of data are applied to make the data as normal-like as possible and, thus, increase the validity of the associated statistical analyses.
Often it can be easier to eschew traditional machine learning algorithms of dealing with time-series data and, instead, use newer, so-called non-linear methods that are not dependent on the distribution of the data.
As a final remark, while all the following transformations and scaling methods can be applied to features directly, an interesting spin with...
