Stepwise linear regression is how they do it. All statistical packages marketed since the late 1970's include the procedures. It's what the AI craze of the ea

Stepwise linear regression is how they do it. All statistical packages marketed since the late 1970's include the procedures. It's what the AI craze of the early 1990's used in their black box apps. Basically you associate a narrow set of outcomes (Filing for divorce, filing for bankruptcy, reporting a stolen credit card) and a mountain of possible predictors (store type, store name, store location, number of purchases from store in the billing cycle, ranges of purchases from a given store within the cycle). The stat program builds a regression equation out of the raw data by literally predicting trying to predict the past from data in the further past. The cool thing is that you can't put in too many variables (as in you don't have to know if something may be related or not) since the uncorrelated variables will simply vanish in the final equation.