Evaluating Real-World Generalizability of Algorithm Selection Models

Algorithm Selection (AS) aims to identify the most suitable optimization algorithm for a given problem using problem characteristics and past performance data. In this study, we examine the generalization of AS models across synthetic and real-world optimization landscapes. We consider two academic benchmark suites (BBOB and CEC) and two real-world problem sets (robotics trajectory optimization and UAV path planning). Through cross-benchmark evaluation, we analyze how AS models transfer between domains, identify where generalization fails, and provide insights into the robustness of current AS approaches.