WG2 seminar: Counterfactual Explanations for regulatory Benchmarking

Title: Counterfactual Explanations for regulatory Benchmarking
Abstract: There is an unprecedented need for transparency and interpretability in Machine Learning, i.e., explain how models arrive at decisions. An effective class of explanations are counterfactuals. In this talk we will apply the concept of counterfactual explanations to regulatory Benchmarking, i.e., find the optimal set of actions that can be taken by an instance such that the model at hand would have assigned a higher efficiency to the firm. We propose a novel Mixed Integer Programming model that minimizes a cost function, measuring the dissimilarity between the given instance and the counterfactual instance, and allows us to incorporate additional constraints like sparsity and actionability. Real-world data has been used to illustrate our model.
Presenter: Miren Jasone Ramirez Ayerbe