Local regression, also known as LOESS or LOWESS, is a non-parametric regression method that fits simple models to localized subsets of the data to build a curve that captures the underlying trend. It is particularly useful for smoothing scatterplots and modeling complex relationships without assuming a specific global functional form.