<p>Feature selection (FS) is the task of identifying the most appropriate subset from all the features by eliminating redundant and irrelevant features. This process can be defined as an optimization process, which aims to reduce the dimensions and enhance the performance of the classification models. In this paper, an improved version of the well-known metaheuristic algorithm called differential evolution (DE) is proposed to perform the FS tasks. The proposed algorithm, referred to as DCDE, is developed to improve convergence speed while mitigating critical issues related to stagnation and premature convergence at local optimal solutions. The diversified clustering and archive strategies are employed to DE to address these challenges, alongside the parameter tuning guided by historical success, the Gaussian and Cauchy distributions. The validation of the developed DCDE algorithm is conducted on the IEEE CEC2017 and IEEE CEC2022 benchmark set of single-objective bound-constrained problems, and later has been extended to its binary version to solve the FS. We have used 22 FS datasets to analyze the efficacy of the proposed DCDE algorithm and compared it with other metaheuristic algorithms. Comparison and analysis of results using several performance metrics have verified the promising and reliable performance of the proposed strategies of the DCDE algorithm to solve continuous and binary FS problems. The DCDE algorithm demonstrated superior performance over all the compared algorithms in more than 60% of the CEC2017 problems, over 70% of the CEC2022 problems, and over 35% of the FS problems.</p>