The machine learning based iterative compilation can effectively predict new programs’ best optimization parameter combination. The existing methods suffer some problems, such as low search efficiency of optimization parameter combination, inappropriate representation of programs features and unsatisfactory prediction accuracy in model training process. The machine learning based iterative compilation is a hotspot research in the field of iterative compilation, and its challenge lies in choosing learning algorithms, optimizing parameter search and program features representation. Based on the supervised learning technique, we propose an optimization parameter predictive method, called SLOPS. We search the optimal parameter space by constraining the multiobjective PSO algorithm, and find the best optimization parameters of the sample function. Then we extract the features of the function through dynamic and static program feature representation techniques. Finally, a supervised learning model is constructed by the KNN and the Softmax regression based on the samples composed of program features and optimization parameters, which is used to predict the optimization parameters of new programs. Experimental results show that the SLOPS can achieve better prediction performance on NPB benchmarks and scientific programs.