Explainability For Efficient And Trusted Decision Optimization

Abstract

A business user of decision optimization models is always interested in their explanations to identify errors/biases in their formulation and further identify of how their business process can be improved. Explanations for decision optimization can be broadly classified into two paradigms (i) explaining optimal solution in terms of decision variables, (ii) explaining the optimal solution in terms of problem-specific parameters. Both paradigms can be explained by learning interpretable surrogate models, however, we also focus on methods which explain decisions using sensible alternate solution or identify critical constraints in solution.