Aprendizado das preferências do decisor usando aprendizado de máquina em problemas multicritério

• Ano de defesa: 2023
• Autor(a) principal: Marcos Antonio Alves
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA ELÉTRICA; Programa de Pós-Graduação em Engenharia Elétrica; UFMG
• Palavras-chave em Português: Tomada de decisão escalável; Matrizes pareadas; Tomada de decisão multicritério; Aprendizado em Batches; Análise hierárquica de processos; Engenharia elétrica; Processo decisório; Matrizes (Matemática); Algoritmos; Aprendizado do computador; Métodos numéricos; Processo decisório por critério múltiplo
• Link de acesso: http://hdl.handle.net/1843/58355

The Analytic Hierarchy Process (AHP) multicriteria method can be cognitively demanding for large-scale decision problems due to the need for the decision-maker to make pairwise comparisons among all the available alternatives. To address this issue, in this thesis we propose an interactive method that uses batch learning to provide scalability for classical AHP, called Scalable AHP. The Scalable AHP involves a machine learning algorithm that learns the decision maker's preferences through evaluations of small subsets of solutions and guides the search for the optimal one. The methodology was tested on different optimization problems, artificial and real ones, with different dimensions and Pareto surfaces in order to validate the applicability of the proposal. A one-factor-at-a-time experimentation of each hyperparameter was performed, from evaluating the number of alternatives to be presented to the decision maker, the most suitable machine learning method for each problem, as well as strategies for selecting and recommending solutions in the iterative process. The results demonstrate that the Scalable AHP is capable of learning the utility function that characterizes the decision maker in approximately 15 iterations with only a few comparisons, resulting in significant savings in time and cognitive effort. The initial subset of alternatives can be chosen randomly or following some clustering strategy. Subsequent alternatives are recommended based on some distance metric throughout the iterative process, with the best selection strategy depending on the type of problem. Recommendation based solely on the smallest Euclidean or Cosine distances reveals better results on linear problems. The proposed methodology can also easily incorporate new parameters and multicriteria methods based on pairwised comparisons.