Genetic algorithms (GAs) are approximate solving methods that have been originally proposed to achieve unconstrained optimization. To handle constrained problems, which is the case for the majority of real-life circumstances, GAs must be equipped with a constraint-handling mechanism. Transformation functions (TFs) are among the constraint-handling approaches that intervene in the phenotypic space. In this paper, we study the impact of considering constraint priorities on the GA performance when it deals with heavily constrained problems. Priorities are set by integrating a constraint order into the TF definition. We consider different TF forms enhanced with a fuzzy inference engine to find the best constraint ordering. Finally, we conduct an experimental study to assess the performance of the proposed approach on the semi-supervised graph partitioning problem. The obtained results show with statistical evidence that the proposed fuzzy method is promising.