Abstract: Selecting frequent subgraph as feature in graph datasets classification based on frequent subgraph plays a very important role. hhe new feature selection strategy was presented based on the information about classes,which is to select unictue and distinctive frectuent subgraphs as feature subgraphs from the candidate subgraphs. The results showed:in the compound data classification, the selection strategy is superior to AutoSVM's and CEP's in classification performance.

Key words: Frequent subgraph pattern, Graph classification, Uraph mining, Feature selection