Optimized Class Association Rule Mining using Genetic Network Programming with Automatic Termination

Association rule mining is one of the tasks of data mining and it has been extensively studied during the last years. As a consequence, recently, several methods for extracting association rules have been developed. Some methods use Evolutionary Algorithms to extract association rules. Among them, a relatively new method using Genetic Network Programming (GNP) has been developed and its effectiveness has been shown, which outperforms other conventional algorithms. However, there still remain some issues mainly focused on performance. To improve the conventional GNP data mining algorithmic efficiency without loss of reliability, a GNP enhanced with an automatic termination criteria named AT-GNP is proposed in this paper. Indeed, in an effort to save computational resources, the objective is to stop the search right before unnecessary function evaluations are performed. The concept of Gene Matrix (GM) is used to direct the search and to stop it at a proper instant. An extensive comparison between the conventional GNPbased association rule mining and AT-GNP is performed in the simulations to evaluate the performance. Finally, the association rules extracted using both methods are applied to the classification problems and the prediction accuracies of them are compared with other conventional approaches. Keywords-Association rule mining; classification; evolutionary computation; genetic network programming; termination criteria.