Genetic algorithm (GA) is a search heuristic that is commonly used for feature selection. The main drawback of GA lies in its unstable results for a random initialization population and background correlation; robust results can only be obtained through a series of runs. This paper proposes the use of selected frequency curve (SFC) analysis to evaluate variable importance based on the results of a classical GA. Three statistical parameters are proposed for the quantitative definition of variable importance based on the SFC. The proposed method was applied to three benchmarking datasets obtained from previous works. This was done in conjunction with the use of different regression and classification methods, and the results were compared with those of a classical GA. The results revealed the robustness and superiority of the combination of GA and SFC analyses (GA-SFC) compared with the use of classical GA. •A robust method, SFC analysis, was proposed for variable selection.•SFC analysis is based on genetic algorithm (GA) but prior to it.•Variable importance could be got in one run GA base on SFC analysis.•Variable importance could be reflected by three statistics based on SFC.