A1043
Title: Doubly robust counterfactual classification
Authors: Kwangho Kim - Korea University (Korea, South) [presenting]
Edward Kennedy - Carnegie Mellon University (United States)
Jose Zubizarreta - Harvard University (United States)
Abstract: Counterfactual classification is studied as a new tool for decision-making under hypothetical (contrary to fact) scenarios. A doubly robust nonparametric estimator is proposed for a general counterfactual classifier, where flexible constraints can be incorporated by casting the classification problem as a nonlinear mathematical program involving counterfactuals. Next, the rates of convergence of the estimator are analyzed, and a closed-form expression is provided for its asymptotic distribution. The analysis shows that the proposed estimator is robust against nuisance model misspecification and can attain fast root-n rates with tractable inference even when using nonparametric machine learning approaches. The empirical performance of the methods is studied by simulation and application on recidivism risk prediction.
