Logo PUC-Rio Logo Maxwell
ETDs @PUC-Rio
Estatística
Título: INTEGRATED ESTIMATE-AND-OPTIMIZE DECISION TREES LEARNING FOR TWO-STAGE LINEAR DECISION-MAKING PROBLEMS
Autor: RAFAELA MOREIRA DE AZEVEDO RIBEIRO
Colaborador(es): BRUNO FANZERES DOS SANTOS - Orientador
Catalogação: 10/JUL/2025 Língua(s): ENGLISH - UNITED STATES
Tipo: TEXT Subtipo: THESIS
Notas: [pt] Todos os dados constantes dos documentos são de inteira responsabilidade de seus autores. Os dados utilizados nas descrições dos documentos estão em conformidade com os sistemas da administração da PUC-Rio.
[en] All data contained in the documents are the sole responsibility of the authors. The data used in the descriptions of the documents are in conformity with the systems of the administration of PUC-Rio.
Referência(s): [pt] https://www.maxwell.vrac.puc-rio.br/projetosEspeciais/ETDs/consultas/conteudo.php?strSecao=resultado&nrSeq=71509&idi=1
[en] https://www.maxwell.vrac.puc-rio.br/projetosEspeciais/ETDs/consultas/conteudo.php?strSecao=resultado&nrSeq=71509&idi=2
DOI: https://doi.org/10.17771/PUCRio.acad.71509
Resumo:
Several decision-making under uncertainty problems found in industry and studied by the scientific community can be framed as a two-stage stochastic program. In the past decades, the standard framework to address this class of mathematical programming problems followed a sequential two-step process, usually referred to as estimate-then-optimize. Firstly, a predictive distribution of the uncertain parameters is estimated, based on some machine/statistical learning (M/SL) method. Then, a decision is prescribed by solving a surrogate of the two-stage stochastic program using the estimated distribution. In this context, most M/SL methods typically focus only on minimizing the prediction error of the uncertain parameters, not accounting for its impact on the underlying decision problem. However, practitioners argue that their main interest is to obtain near-optimal solutions from the available data with minimum decision error rather than a least-error prediction. Therefore, in this work, we discuss a new framework for integrating prediction and prescription into the predictive distribution estimation process to be subsequently used to devise a decision to be implemented. We particularly focus on decision trees and study decision-making problems representable as two-stage linear programs. Firstly, we propose a workable framework to account for the impact of the decision-making problem dynamics on the predictive distribution estimation process. A non-convex mathematical programming problem is formulated to characterize the integrated prediction- and prescription-oriented estimation problem. Then, through a set of reformulation procedures, we recast the non-convex mathematical program as a Mixed-Integer Programming (MIP) problem. Acknowledging the difficulty of the MIP reformulation to scale to mid- and large-scale instances, we devise a computationally efficient recursive-partitioning heuristic strategy for the integrated prediction- and prescription-oriented estimation problem leveraging the structure intrinsic to decision trees. A key feature of the proposed decision-making framework is its instant decision assessment capability. When a new context arises, generating a prescription reduces to identifying the corresponding leaf it falls into and retrieving the precomputed solution of the associated two-stage problem. This enables fast, direct prescriptions without the need for a time-consuming optimization step. A set of numerical experiments is conducted to illustrate the capability and effectiveness of the proposed framework using three distinct two-stage decision-making problems. We benchmark the proposed approach against prescriptions devised by various alternative frameworks. We consider five predict/estimate-then-optimize benchmarks that rely on commonly used predictive and distribution estimation methods found in the literature and three benchmarks based on integrated predict-and-optimize decision-making processes, discussing the benchmark s solution quality and the computational capability of the MIP reformulation.
Descrição: Arquivo:   
COMPLETE PDF