Discretization of multidimensional attributes can improve the training speed and accuracy of machine learning algorithm. At present, the discretization algorithms perform at a lower level, and most of them are single attribute discretization algorithm, ignoring the potential association between attributes. Based on this, we proposed a discretization algorithm based on forest optimization and rough set (FORDA) in this paper. To solve the problem of discretization of multi-dimensional attributes, the algorithm designs the appropriate value function according to the variable precision rough set theory, and then constructs the forest optimization network and iteratively searches for the optimal subset of breakpoints. The experimental results on the UCI datasets show that:compared with the current mainstream discretization algorithms, the algorithm can avoid local optimization, significantly improve the classification accuracy of the SVM classifier, and its discretization performance is better, which verifies the effectiveness of the algorithm.