Feasible Regions Recovery of Optimization Models

Many real decisions are constrained in ways that are only partly written down. A clinician may accept one treatment plan and reject another, a dietitian may approve some meal plans as “realistic” and flag others as not workable, or an operations team may keep only solutions that fit staffing rules and safety checks. In all of these settings, the true constraints live in a mix of guidelines, habits, and expert judgment. This research stream aims to find the hidden feasible region from past examples, so that we can encode it as an optimization model that produces decisions experts are likely to accept. The approach uses inverse optimization to infer the boundary of what is allowed. It learns constraints that make historical accepted decisions feasible, and it can also use rejected decisions to sharpen the boundary and avoid repeating known bad options. The result is a recovered feasible region that can be used as a practical engine for decision support. It provides a transparent way to screen new candidate decisions as plausible or not, and it supports “what if” reasoning about how recommendations would change when priorities shift.