This chapter introduces the Racos (RAndomized COordinate Shrinking) optimization algorithm, a novel approach designed to address complex optimization problems in both continuous and discrete search spaces. Building on the theoretical insights from the previous chapter, Racos minimizes critical factors such as error-target dependence and shrinking rate to enhance optimization efficiency. The algorithm integrates a randomized coordinate shrinking classification technique, which effectively balances exploration and exploitation in the search process. The chapter is structured as follows: Sect. 5.1 details the Racos algorithm, Sect. 5.2 presents empirical evaluations on benchmark functions, Sect. 5.3 applies Racos to spectral clustering tasks, and Sect. 5.4 examines its performance in classification tasks using Ramp loss. Experimental results demonstrate Racos’s superiority over state-of-the-art derivative-free optimization methods, highlighting its scalability, robustness, and effectiveness in high-dimensional and complex optimization scenarios.