Machine-DM: A Machine-Based Decision-Maker for Benchmarking Interactive MCDM Procedures
Abstract
Interactive multi-criterion decision-making (iM-CDM) procedures typically require a human decision-maker (DM) to iteratively compute a preferred Pareto-optimal solution by incorporating her/his preference information. Many iMCDM procedures involve classifying the objectives according to whether they need to be improved, relaxed, kept unchanged, or allowed to vary freely from its current value. Based on this classification, the DM also specifies bounding parameters in the form of upper and lower limits on objectives for the current solution. Next, a singleobjective Achievement Scalarization Function (ASF) is often formulated and solved to compute a preferred Pareto-optimal (PO) solution. DM provides preference information, computes progressively more preferred solutions iteratively, compares them to finally identify the most desirable one. Despite the rich body of literature in evolutionary multi-objective optimization (EMO) and iMCDM, there is a lack of benchmarking iMCDM procedures due to the inherent involvement of a human DM. In this paper, we propose a Machine-based Decision Maker (MachineDM) that employs pre-trained machine learning (ML) models to perform the iMCDM tasks, thereby eliminating the dependence on human inputs. We further integrate the Machine-DM into an existing iMCDM procedure -Step Method (STEM)- and propose a Machine-DM based STEM (MachDM-STEM) procedure to illustrate the working of Machine-DM approach. The application of MachDM-STEM to both benchmark and realworld problems demonstrates the strong generalization capability of Machine-DM and highlights its potential to open new avenues for algorithmically benchmarking a broader class of iMCDM procedures, thereby advancing the combined EMO-MCDM field.
Authors: Deepanshu Yadav, Kalyanmoy Deb
Published in: International Conference on Emerging Techniques in Computational Intelligence (ICETCI) (2026)