Simple Prediction Methods

Abstract

Predictive modeling has gotten sophisticated over the years. It is a major discipline of data science and includes some of the most sophisticated mathematics and statistical engineering found anywhere. That being said, some of the tools for prediction are straightforward. We can start with the package prophet, graciously given to the world by Facebook. If you compare prophet to some of the older time series prediction methods, you will be impressed with its simplicity. It has all the trending, seasonality, and other mathematical patterns of older systems but does not force you to “get involved.” You can just enter a dataframe with one column of dates and another column of some numeric value (birds per square mile, prisoners in Alabama, rain forest size, etc.) and then predict future values. The most time-consuming part is getting the dates and column headers in the format prophet requires. After prophet, I’ll present an older method, Holt-Winters (time series) and multivariate regression, where one or more variables (predictors) are used to estimate some variable of interest, termed the response.