Metaphor-based algorithms for learning the preference model parameters of FlowSort from large sets of assignment examples
[ 1 ] Instytut Informatyki, Wydział Informatyki i Telekomunikacji, Politechnika Poznańska | [ S ] student | [ P ] employee
2023
chapter in monograph / paper
english
- Multiple Criteria Decision Analysis
- Preference learning
- Evolutionary algorithm
- Mathematical programming
- Assignment examples
EN We consider the problem of deriving parameters of the preference model employed in the multiple criteria sorting method called FlowSort. We propose a suite of preference learning algorithms based on differential evolution and simulated annealing, their combinations with mathematical programming, and a dedicated heuristic. They are tested on various monotonic benchmark datasets and compared in terms of 0/1 loss. The evolutionary algorithm and the dedicated heuristic prove competitive against state-of-the-art preference learning methods. The former attains better results when coupled with boundary profiles for all considered datasets. For other methods, there is no clear indication that using the class limits is more advantageous than class prototypes.
24.07.2023
283 - 286
20
140