Large-Scale Association Rule Discovery from Heterogeneous Databases with Missing Values using Genetic Network Programming.
Eloy Gonzales, Takafumi Nakanishiand, Koji Zettsu · 2011
Association Rule Mining is an important data mining task and it has been studied from different perspectives. Recently multi-relational rule mining algorithms have been developed due to many real-world applications. However, current work has generally assumed that all the needed data to build an accurate model resides in a single database. Many practical settings, however, require the combination of tuples from multiple databases to obtain enough information to build appropriate models for extracting association rules. Such databases are often autonomous and heterogeneous in their schemes and data. In this paper, a method for association rule mining from large, heterogeneous and incomplete databases is proposed using an evolutionary method named Genetic Network Programming (GNP). Some other association rule mining methods can not handle incomplete data directly. GNP uses direct graph structure and is able to extract rules without generating frequent itemsets. The performance of the method is evaluated using real scientific heterogeneous databases with a high rate of missing data.