Mining and representing rare association rules through the use of genetic programming

Whereas the extraction of frequent patterns has focused the major researches in association rule mining, the requirements of reliable rules that do not frequently appear is taking an increasing interest in a great number of areas. This field has not been explored in depth and most algorithms for mining infrequent association rules follow an exhaustive search methodology, which hampers the extracting process because of the size of the datasets. The importance of discovering patterns that do not frequently appear in a dataset and the promising results obtained when using evolutionary proposals in the field of frequent pattern mining motivates the evolutionary proposal for discovering rare association rules presented in this paper. Here, a context-free grammar is described and applied to adapt individuals to each particular problem or domain. The use of both an evolutionary approach and a context-free grammar reduces the memory requirements and provides the possibility of extracting any kind of rules, respectively. The experimental study shows that this proposal obtains a set of reliable infrequent rules in a short period of time.