A new proposal for graph classification using frequent geometric subgraphs

Geometric graph mining has bees identified as a need in many applications. This technique detect patterns with some tolerance under a geometric transformation. To meet this need, some graph miners have been developed for detecting frequent geometric subgraphs. However, there are few works for applying this kind of geometric patterns as feature for classification tasks. In this paper, a new geometric graph miner and a framework for using frequent geometric subgraphs in classification, are proposed. Our solution was tested in two real collections. The experimentation on these collections shows that our proposal gets better results than graph-based image classification using non-geometric graph miners.