Evolutionary Computations (ECs) are powerful search algorithms for nonlinear problems, and have been. widely studied. Generally, calculation of EC to acquire expected solutions takes much time because they need repeated calculation for search solutions. Hanaki et al. proposed the fitness inference to reduce evaluation time by simplifying calculation of the fitness of chromosomes. In fitness inference, how chromosome similarities is defined is very important corresponding to that of solutions. We studied chromosome similarities considering the solution similarities, and propose new chromosome similarities with the weight on each locus determined by the change of information on the locus using Sequential Difference Fitness Value Allocation. We used benchmark functions to study the feasibility of the proposed method and found that effective weights on loci suitable for the search space are automatically generated, and that the proposed method enables effective fitness inference.