For a retail sales forecast, the mean of an SKU’s sales over the previous N days is a candidate feature. It is the arithmetic average: add the sales values and divide by the number of values. This gives the model one summary of recent sales, alongside candidates such as the sum, minimum, maximum, and standard deviation. section 11.5.1
A useful summary is not automatically a useful feature. Test whether the mean improves forecasts, then weigh any gain against interpretability, computation time, and feature stability. section 11.5.1
