Improving heuristic regression analysis

Authors:
Floyd A. Miller
Affiliations:
The Glidden Company, Jacksonville, Florida
Venue:
ACM-SE 6 Proceedings of the 6th annual Southeastern regional meeting of the Association for Computing Machinery and national meeting of Biomedical Computing - Volume 1
Year:
1967

Citing 2
Cited 2

Robot data screening: a solution to multivariate type problems in the biological and social sciences

Communications of the ACM
Computers and Thought

Computers and Thought

Experiments with the M & N tree-searching program

Communications of the ACM
Experiments in automatic learning for a multipurpose hueristic program

Communications of the ACM

Quantified Score

Hi-index	0.03

Visualization

Abstract

Heuristic Regression Analysis, a new probabilistic predictor selection concept that allows a computer to automatically "learn" the best regression model, has become a practical and economical tool for profit-oriented industry replacing the usual stepwise regression approach. Wide experience with this computer substitute of human problem solving effort has led to substantial improvements of the technique. Starting with forty variables, simple models having three, four or five predictor terms are rapidly located. An "editor" routine, which has been developed to printout only the best three models generated during the learning iterations, significantly reduces the time required by the user to analyse the final models.