Enhancing the interpretability of GPLSIMs through Counterfactual Explanations

E. Carrizosa, M. Alonso-Pena, A. Navas-Orozco

Generalized Partially Linear Single-Index Models (GPLSIMs) are a wide class of semiparametric statistical models that generalize many simpler well-known models, namely generalized linear models (including linear and logistic regression), (generalized) single-index models, and partially linear single-index models. They offer great flexibility, but their interpretability is limited. In this work, we address this limitation by leveraging Counterfactual Explanations (CEs). CEs intuitively explain the outcome of a given individual. The CE of a record is a vector, close enough to the record, with maximal outcome. Their computation for GPLSIMs is expressed as a nonconvex piecewise polynomial optimization problem, globally solvable using state-of-the-art solvers like Gurobi. Finally, we present CounterfactUS, an open source Python package that integrates GPLSIM fitting with CE generation, enhancing the transparency of these semiparametric models without sacrificing flexibility and performance.

Palabras clave: Data-driven decision making, Counterfactual Explanations, Generalized Partially Linear Single-Index Models, Semiparametric statistics, Interpretability

GT AMyC III: Mathematical Optimization for Transparent Decision Making
4 de septiembre de 2026  15:30
Aula 28

E. Carrizosa, J. C. Castro Gómez, V. Guerrero

E. Carrizosa, R. DE LEONE, M. MAGAGNINI

J. Martín-Chávez, E. Carrizosa, C. Molero-Río

C. Molero-Río, C. D'Ambrosio