Subgraph mining in graph-based data using multiobjective evolutionary programming

This work proposes multiobjective subgraph mining in graph-based data using multiobjective evolutionary programming (MOEP). A mined subgraph is defined by two objectives, support and size. These objectives are conflicting as a subgraph with high support value is usually of small size and vice-versa. MOEP applies NSGA-II's nondominated sorting procedure to evolve the population during the subgraph generation process. An experimental study on five synthetic and real-life graph-based datasets shows that MOEP outperforms Subdue-based methods, a well-known heuristic search approach for subgraph discovery in data mining community. The comparison is done using hypervolume, C and Ie multiobjective performance metrics.