Design of an expert system for regression model building

Abstract

It is often the case in research that an experimenter has data for some large superset of p independent variables (predictors) in a regression model and believes that a subset of these predictors may be entirely adequate for his purpose. RMEX is a statistical expert system that provides the user with the strategy for selecting the "best" regression equation in multiple regression analysis. Currently existing softwares assume that the user has considerable amount of statistical knowledge. On the other hand, RMEX explains necessary statistical concepts and procedures step by step for doing linear regression. In the preliminary analysis, a classification of the various types of multivariate techniques is presented and general guidelines for their applications are provided. Next, the selection of a "best" subset of predictors is guided explaining the intermediate and final results. RMEX is implemented in INSIGHT 2(shell) and interfaces with pascal programs.