Graph Classification using Co-occurrent Frequent Subgraphs

Frequent subgraphs are widely used as feature vectors in graph classification. It is very important for a graph classification performance to select useful frequent subgraphs from many mined frequent subgraphs. The existing feature selection studies have a shortcoming that is a classification performance degradation from the lack of discrimination power among individual patterns. In this paper, we propose a model based search tree using co-occurrence of frequent subgraphs, and suggest an efficient algorithm. The proposed approach selects more discriminative frequent features considering both discriminative individual and discriminative co- occurrent frequent subgraphs. In experiment, we show that our proposed technique can have a higher graph classification performance compared to existing approach.