<p>Subspace clustering on high-dimensional data usually suffers from correlated and contaminated features, which seriously restrict the subspace distribution and graph embedding procedure in practical applications. Therefore, how to preserve intrinsic structure accurately on robust subspaces still needs to be further explored. In this paper, we propose a robust Graph Embedded Subspace Clustering model with Entropy-based Feature Weighting (GSCEFW) to differentiate feature weights and substantially facilitate manifold preserving during subspace learning. In particular, an optimal graph exploration term guided by pseudo-label learning is introduced to subspace clustering framework, which imposes dual-structural constraint on subspace representation to strengthen its block diagonal contour. Then, an entropy-based feature weighting term is considered to automatically mitigate the adverse effect from noisy or irrelevant features during data reconstruction. Finally, an alternative optimization method is developed to solve the challenging objective function, together with theoretical algorithm analysis. Extensive experiments on benchmark datasets demonstrate the effectiveness and superiority of the proposed GSCEFW model compared with the state-of-the-art models.</p>