Linear programming system identification

We define a version of the Inverse Linear Programming problem that we call Linear Programming System Identification. This version of the problem seeks to identify both the objective function coefficient vector and the constraint matrix of a linear programming problem that best fits a set of observed vector pairs. One vector is that of actual decisions that we call outputs. These are regarded as approximations of optimal decision vectors. The other vector consists of the inputs or resources actually used to produce the corresponding outputs. We propose an algorithm for approximating the maximum likelihood solution. The major limitation of the method is the computation of exact volumes of convex polytopes. A numerical illustration is given for simulated data.