Stochastic search methods are widely used when it comes to design synthesis and optimization of response-based objective functions. In engineering applications, the objective function is typically expensive to evaluate, and stochastic search methods lack efficiency, resulting in the necessity of extensive design evaluations. In order to improve stochastic search methods, we propose a Machine Learning (ML)-based augmentation, consisting of three modules: a design archiver, a data modeler, and a modification advisor. These three modules cooperatively work together to store the gathered data during the design process, build up a representative model of the observations made, and advise the search for further sequences of modifications to apply. The proposed method is benchmarked against its unaugmented parent method in placing cooling channels in a die casting mold. The results show that the efficiency of the method is significantly improved when augmented with ML, i.e. similar results are obtained with 25–50% fewer evaluations. Additionally, the robustness and reliability of the optimization process is improved with a standard deviation of the obtained results that is 60–85% smaller. It is shown that the search strategy can be significantly improved with the proposed method, resulting in shorter running times and more reliable convergence behavior.