A feature selection method of a forest optimization algorithm based on a scoring mechanism belongs to a new evolutionary algorithm, and allows a solution with high fitness value to have a higher probability of being selected from the point of view of a candidate set selection strategy and a scoring strategy of a feature set by a single feature. In this way, each choice is not completely random, but with a certain tendency. Therefore, the probability of falling into a local optimal solution can be reduced, and an approximate optimal subset of features can be quickly and effectively searched. The feature selection method can be applied in the social network classification, web text data mining, image processing and other high-dimensional data fields, and can play a very good role in dimensionality reduction.