SBIR Phase I: Automating Regression: An Optimization Approach

The Small Business Innovation Phase I Research will develop a formal algorithm, and a corresponding software package, which constructs and evaluates regression models automatically and can handle large data sets efficiently. Regression analysis is widely used in all areas of physical and social sciences, medicine, management, and engineering. Despite its widespread use, there is a significant element of art in building regression models. In addition, the wide availability of data makes it important to be able to perform regressions with very large data sets and a large number of potential explanatory variables. The objective is to take the art out of the construction of regression models, and, in doing so, will achieve better regression models that are significantly faster, and extend their capabilities to build much larger models in the presence of very large data sets. The commercial benefits of an automated regression package would allow data mining in diverse areas such as marketing, health care, insurance, credit cards, and finance. An automated regression package that can handle very large data sets and many explanatory variables can become the critical engine for data mining in all these industries.