UNSUPERVISED FEATURE SELECTION METHOD AND DEVICE

摘要

Disclosed are an unsupervised feature selection method and device. In the method, a feature topology diagram is firstly constructed according to the similarity between features, and then the feature topology diagram is divided, so that feature nodes with a relatively high similarity are divided into the same connected graph to realize the feature clustering of sample data, and thus the features of the sample data can also be selected in a scenario where a classification mark of the sample data cannot be pre-determined; after the feature clustering is completed through the connected graph, a node can be further selected from each connected graph, and the features corresponding to the node are written into a target feature set as representative features, so as to obtain comprehensive and non-repetitive representative features corresponding to the entire sample data. Therefore, in the present application, without needing to depend on a classification mark of sample data, unsupervised feature selection can be achieved, and it is ensured that two or more similar features do not appear in a target feature set, so that the target feature set can describe the sample data more intuitively, thereby avoiding information redundancy.