- Article
Bio-inspired continuous metaheuristics are routinely adapted to binary combinatorial optimization through binarization schemes that pair a transfer function with a discretization rule, often treated as problem-agnostic transformations. Recent evidence on emergent swarm behavior suggests that the behavior observed during optimization depends on the interaction between algorithmic mechanisms and the problem landscape. This raises the question of whether the relative effectiveness of binarization schemes also changes across structurally different binary problems. Building on previous evidence from the Set Covering Problem, where Elitist-based discretization rules showed the strongest performance, we examine this question using the 0–1 Knapsack Problem as a structurally distinct problem. Sixteen transfer functions (S-shaped, V-shaped, X-shaped and Z-shaped) were combined with Standard, Elitist, and Complementary discretization rules, yielding 48 binarization schemes embedded in Particle Swarm Optimization and Grey Wolf Optimizer and evaluated on 21 Pisinger instances over 31 independent runs. All schemes were compared under the same fixed, low computational budget, using multiplicity-corrected post hoc tests (Nemenyi and Holm) and Vargha–Delaney effect sizes. The results show a different performance hierarchy from that previously reported for Set Covering. With both algorithms, the Standard rule produced the best-ranked configurations (S-STD with GWO; V-STD followed by S-STD with PSO), and every Elitist configuration was significantly worse than S-STD. Within the S-shaped family, S1-STD significantly outperformed S2-STD, S3-STD and S4-STD in most instances with 500 or more items, with large effect sizes. The ordering was identical across the three Pisinger correlation classes. Because the Set Covering evidence comes from a previous study with a different protocol, the contrast is interpreted as evidence that binarization hierarchies are not transferable between problems, not as proof that problem structure causes the change. These findings motivate Landscape-Aware Binarization (LAB) as a research perspective for studying problem-dependent binarization and identify a unified multi-problem, multi-budget design as the necessary next step.
20 November 2026






















![Overview of the different housing structures according to Drexler et al. [25] by model year (a) and an evaluation of the distribution of the electric drive unit’s weight in kilograms (b).](https://mdpi-res.com/energies/energies-19-05324/article_deploy/html/images/energies-19-05324-g001-550.jpg)




