Although various optimization methods exist in machining, the predominant challenge in practical applications often arises from the experimental effort required to generate sufficient data for reliable models. This work provides an experimental evaluation of the Metamodel-Based Evolutionary Optimizer (MEVO) as a data-efficient alternative for machining optimization and benchmarks its performance against the conventional Response Surface Methodology (RSM) desirability workflow. Slot milling experiments were performed on Aluminum 6061-T651 utilizing three cutting parameters as inputs: axial depth of cut \(a_p\) (mm), feed per tooth \(f_z\) (mm/tooth), and cutting speed \(v_c\) (m/min). The targeted responses were Active Specific Cutting Energy (ASCE, J/mm \(^3\) ) and average surface roughness ( \(R_a\) , \(\mu \text {m}\) ). To address the multi-response nature of the problem within a single-objective optimization framework, both responses were mapped to individual desirability functions and subsequently aggregated into a composite desirability D, which was maximized. The RSM approach employed a standard rotatable central composite design with quadratic response models. In contrast, MEVO utilized an initial Latin hypercube sampling to create a minimal training set and iteratively proposed new experiments based on a continuously updated surrogate model. Using best-so-far stagnation stopping criteria, MEVO achieved composite desirabilities of 0.8932 ( \(k=2\) , 9 runs) and 0.8978 ( \(k=3\) , 10 runs), outperforming the RSM solution (0.8363, 21 runs) while significantly reducing the experimental workload by approximately 52-57%. These results demonstrate that MEVO is capable of delivering comparable or superior machining optimization solutions with substantially fewer physical experiments.
