In many real-world problems such as gene selection which is a high dimensional problem, the large number of features is the main challenge. Exhaustive search to find the optimal feature subset is not feasible in a reasonable time and might reduce the performance of the classifier. To address this issue, many binary metaheuristic algorithms are proposed to approximate the optimal solution by removing irrelevant features within an acceptable computational time. This paper presents a comparative analysis to evaluate the efficiency of transfer function-based binary metaheuristic algorithms for feature selection. We compare the performance of popular algorithms Binary Bat Algorithm (BBA), Binary Gravitational Search Algorithm (BGSA) and binary Grey Wolf Optimization (bGWO) with different transfer functions. The experimental results on seven well-known datasets including high dimensional datasets Colon and Leukemia demonstrate that BGSA is an efficient and suitable algorithm for feature selection from both kinds of datasets with low and high dimensions.