In batch process, the change in a recipe will cause the step interference in the system. And the recipe clustering can reduce the influences of the fixed-parameter controller and software instrument. According to the recipe data with different large-scale and high-dimensions, an improved subspace clustering algorithm (ISCA) is put forward in this paper. In order to classify the recipe data, the most similarity of subspace clustering which is applied to determine the search direction of the subspaces is defined in ISCA. Then use IFCA to cluster for the reaction unit of specific recipe model. The simulation results show that ISCA can improve the accuracy and flexibility of the subspace clustering, compared with the traditional high-dimensional subspace clustering algorithm. The ISCA is more suitable clustering for large-scale and high-dimensional recipe data.