Ohnishi KKoga DTIAN-LI YU2023-06-092023-06-092021https://www.scopus.com/inward/record.uri?eid=2-s2.0-85125773435&doi=10.1109%2fSSCI50451.2021.9660010&partnerID=40&md5=48c689d1f00b4cbc4ae28441616286d7https://scholars.lib.ntu.edu.tw/handle/123456789/632315This paper proposes a new test problem for model-based GAs (MBGAs) called the overlapping chains problem (OCP). The investigation of overlapping linkages has been an important issue in MBGAs that adopt linkage detection. Like some other existing test problems, OCP consists of sub-problems with overlapping linkages. Nevertheless, unlike others, the bits used for fitness calculation in OCP are uniquely designed such that they depend on the pattern of the solution candidate. The experimental analysis shows that if the Hamming distance between two randomly generated solution candidates increases, the difference in the bits used for their fitness calculations increases. The analysis also shows that the Hamming distance between two solution candidates with the same or similar high fitness values is quite large and the bits used for the fitness calculation vary slightly even between two such good ones. These results suggest that at each search stage, the bits used for fitness calculation are not completely fixed and that the linkages are difficult to identify. Several state-of-the-art MBGAs are then tested on OCP. Empirical results suggest that OCP provides a new aspect of difficulty for MBGAs to correctly identify linkages and generate high-quality solutions, applying trial-and-error methods to existing ones is essential. © 2021 IEEE.Linkage identification; Model-based genetic algorithm; Overlapping linkages; Test problemHamming distance; Health; Bit patterns; Experimental analysis; Fitness calculation; Fitness values; Linkage identification; Model-based genetic algorithm; Model-based OPC; Overlapping linkage; Sub-problems; Test problem; Genetic algorithmsconference paper10.1109/SSCI50451.2021.96600102-s2.0-85125773435