Impact of Number of Interactions, Different Interaction Patterns, and Human Inconsistencies on Some Hybrid Evolutionary Multiobjective Optimization Algorithms


We investigate the impact of the number of human-computer interactions, different interaction patterns, and human inconsistencies in decision maker responses on the convergence of an interactive, evolutionary multiobjective algorithm recently developed by the authors. In our context an interaction means choosing the best and worst solutions among a sample of six solutions. By interaction patterns we refer to whether preference questioning is more front-, center-, rear-, or edge-loaded. As test problems we use two- to four-objective knapsack problems, multicriteria scheduling problems, and multiobjective facility location problems. In the tests, two different preference functions are used to represent actual decision maker preferences, linear and Chebyshev. The results indicate that it is possible to obtain solutions that are very good or even nearly optimal with a reasonable number of interactions. The results also indicate that the algorithm is robust to minor inconsistencies in decision maker responses. There is also surprising robustness toward different patterns of interaction with the decision maker. The results are of interest to the evolutionary multiobjective (EMO) community actively developing hybrid interactive EMO approaches.