<p>With the increasing complexity of real-world engineering applications, expensive multi-objective optimization problems (EMOPs) have become prevalent. To alleviate the high computational cost associated with EMOPs, various surrogate-assisted evolutionary algorithms (SAEAs) have been proposed. However, most existing SAEAs exhibit performance degradation when applied to high-dimensional problems due to difficulties in surrogate modeling and optimization scalability. To address this challenge, we propose a novel SAEA called GSAEA (Grouping via Sensitivity Analysis Evolutionary Algorithm). Specifically, GSAEA employs sensitivity analysis and difference in contributions to group decision variables into convergence-related variables and diversity-related variables so as to perform targeted optimization for each subset. Furthermore, to further improve the accuracy of surrogate modeling, it constructs multiple radial basis function models to handle different optimization tasks separately. Finally, an infill sampling criterion selects the most promising candidate solutions for real functional evaluations. Experimental results on benchmark test suites containing up to 200 decision variables as well as on a high-dimensional real-world problem show that GSAEA outperforms several state-of-the-art SAEAs in terms of convergence and computational efficiency.</p>