Feature Selection (FS) is an important aspect of knowledge extraction as it helps to reduce dimensionality of data. Among the numerous FS algorithms proposed over the years, Gravitational Search Algorithm (GSA) is a popular one which has been applied to various domains. However, GSA suffers from the problem of pre-mature convergence which affects exploration leading to performance degradation. To aid exploration, in the present work, we use a clustering technique in order to make the initial population distributed over the entire feature space and to increase the inclusion of features which are more promising. The proposed method is named Clustering based Population in Binary GSA (CPBGSA). To assess the performance of our proposed model, 20 standard UCI datasets are used, and the results are compared with some contemporary methods. It is observed that CPBGSA outperforms other methods in 12 out of 20 cases in terms of average classification accuracy. The relevant codes of the entire CPBGSA model can be found in the provided link: https://github.com/ManosijGhosh/Clustering-based-Population-in-Binary-GSA.