A Machine-based Decision-maker Toward Benchmarking Interactive Multi-criterion Decision-making Procedures

Abstract

Interactive multi-criterion decision-making (iMCDM) procedures allow decision-makers (DMs) to provide their preferences to create and compare alternative solutions in an iterative manner and finally arrive at the most preferred Pareto-optimal (PO) solution. Some iMCDM procedures require DMs to provide a clear classification of objectives of the current solution into different categories for further improvement, relaxation, indifference, or satisfaction. Other approaches require DMs to compare two or more competing solutions and indicate the most preferred one. Starting with a predefined or random PO solution, iMCDM procedures, in guidance from humans DMs, iteratively generate new and increasingly more preferred PO solutions by solving appropriate scalarized optimization problems. Due to involvement of human DMs, computational optimization researchers (namely, evolutionary multi-criterion optimization (EMO) researchers), despite being active in developing efficient EMO algorithms, have mostly refrained in venturing into proposing new iMCDM procedures. In this paper, we propose a machine-based decision-maker (Machine-DM) in terms of pre-trained machine learning (ML) models to provide decision-making information. We describe our Machine-DM development process in detail and demonstrate its working through a specific iMCDM procedure – NIMBUS. The proposed ML-based Machine-DM is envisaged to be a core part of our immediate future goal on benchmarking iMCDM procedures. It opens new avenues and encourages computationally-oriented researchers for developing new and efficient iMCDM methodologies to further advance the combined EMO-MCDM field.