MACHINE LEARNING FOR SET-IDENTIFIED LINEAR MODELS

Abstract

Abstract. When a researcher only has partial or incomplete knowledge of a model's parameters, a set of feasible values for those parameters can be known in the place of a single point estimate. These models are known as set-identified linear models. Traditional statistical techniques for estimating linear models may not be appropriate in the set-identified scenario since they rely on the assumption that the parameters are completely identified. In this paper, I will describe machine learning techniques that including the lasso set method, support vector regression, and constrained L1 minimization. They can be useful for this field.